MANAGEMENT SCIENCE

Vol. 00, No. 0, Xxxxx 0000, pp. 000–000

[Link]
ISSN 0025-1909, EISSN 1526-5501

Submitted to Management Science

Food Ordering and Delivery: How Platforms and

Restaurants Should Split the Pie
Jaelynn Oh
The University of Utah, [Link]@[Link],

Chloe Kim Glaeser

University of North Carolina, chloe glaeser@[Link],

Xuanming Su
University of Pennsylvania, xuanming@[Link],

Authors are encouraged to submit new Abstract. Food ordering and delivery platforms generate online demand for restau-
papers to INFORMS journals by means rants and deliver food to customers. In return, restaurants pay platforms a commis-
of a style file template, which includes sion, typically a percentage of the order amount. Platforms offer partner restaurants
the journal title. However, use of a tem- the choice of a range of commission rates, rewarding higher commission payments
plate does not certify that the paper with featured display slots and discounted delivery fees, both of which stimulate
has been accepted for publication in demand. Unfortunately, the current environment is grim: platforms scurry to cover
the named journal. INFORMS journal delivery costs while restaurants gripe about excessive commissions. To understand
templates are for the exclusive purpose current practice, we develop a game-theoretic model with a platform and multiple
of submitting to an INFORMS journal restaurants. Our modeling results highlight two existing problems. (1) Platforms, on
and are not intended to be a true repre- their apps/websites, feature restaurants that are located too far away. Since these
sentation of the article’s final published restaurants do not internalize the platform’s delivery costs, they are willing to choose
form. Use of this template to distribute aggressively high commissions to earn featured display. (2) Platforms charge delivery
papers in print or online or to submit fees that are too high and set delivery boundaries that are too narrow. This is because
papers to another non-INFORM publi- they bear the entire burden of delivery but earn only a fraction of food revenues.
cation is prohibited. To solve these problems, we propose a simple fix to existing commission contracts:
beyond sharing food revenue (currently done but at high commission rates), platforms
and restaurants can also split delivery costs and fees (currently not done). We show
that our method attains first-best, i.e., maximizes the total pie shared by platforms and
restaurants. Using data on a representative city, we numerically show that, on average,
our coordinating contract lowers commission rates by 33.3%, lowers delivery fees by
40.4%, increases restaurant profit by 25.0%, increases platform profit in 30.9% of the
markets, and increases total profit by 13.3%. We discuss the characteristics of markets
that enable our coordinating contract to yield a winning outcome for all parties.
Keywords: platforms, restaurants, food delivery, contracts

1
 Oh, Glaeser, and Su: Food Ordering and Delivery
2 Article submitted to Management Science

1. Introduction
The online food delivery market has steadily grown in many countries in the past decade. The recent
COVID-19 pandemic fueled this growth exponentially: lock-downs and social-distancing requirements
shifted consumer habits towards ordering foods via smartphone apps. Consequently, the U.S. restaurant
delivery business grew from $23 billion in 2019 to about $51 billion in 2020 (Oblander and McCarthy,
2021). Despite the boost in sales, food delivery platforms still struggle to make profits (Rana and Haddon,
2021). Among the major food delivery platforms in the U.S., DoorDash is the only company that was prof-
itable for a quarter in 2020, while others have never been profitable over their multiple years of operation
(Yahoo Finance 2021, 2022a, 2022b, 2022c). Food delivery platforms operate on thin margins from restau-
rant food commissions and delivery and service fees charged to consumers, the majority of which are spent
on operating expenses, such as paying delivery drivers. As the pandemic wanes, food delivery platforms are
working hard to find a sustainable business model.
The recent controversy over food commissions adds another challenge to the platform’s already tight
operating profit margin. Various news media highlighted that given traditionally low profit margins of
restaurant businesses, the typical 20 - 30% commission rate charged by platforms leave restaurants with
little to no profit (Forman, 2019). From the perspective of restauranteurs, giving up a lion’s share of their
profits in return for additional demand streaming from the platform’s website does not appear to be a fair
arrangement. Many municipalities and state legislatures chose to support local restaurants by temporarily
capping what the platforms can charge to restaurants during the pandemic (Rana, 2021). Some city coun-
cils even permanently capped the commissions (Davalos, 2021). While the caps on commissions of food
delivery platforms increase restaurant welfare nominally, Sullivan (2023) finds that they overall decrease
platform and customer welfare. The fact that food delivery platforms rarely make a profit suggests that this
would not be a viable solution to address the low operating margin problem for both parties. In this paper,
we address this debate and seek a mutually beneficial arrangement for restaurants and the platform.
In response to the increasing pushback over commissions, major food delivery platforms have recently
changed their methods for setting the commission rates. Initially, the majority of platforms in the US,
such as Seamless, UberEats, and DoorDash, negotiated fixed percentage commission fees with restaurants
that ranged from 10% to 30% (Littman, 2019). Some of these platforms also negotiated the fixed percent-
age commission rate with restaurants based on order volume (Dickinson, 2023). This single marketplace
fee included the costs of onboarding new delivery couriers, paying couriers, marketing, and overhead. In
2021, Uber Eats, Postmates, DoorDash, and Grubhub, the market leaders in food delivery platforms, intro-
duced tiered pricing structures. There are three options for tiered pricing on major platforms like Grubhub,
UberEats, and DoorDash. The commission rates range from 15% for the basic service plan to 30% for the
“sponsored” service plan. Figure 1 provides a snapshot of these fee structures offered by UberEats.1 We can
1
[Link] Retrieved September 27, 2023
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 3

see that the Lite plan charges 15% in fees, but all customers must also pay a delivery fee to receive an order
from the restaurant. In addition, the restaurant will only be shown to customers when they purposefully
search for it on the platform.
On the other hand, the Plus or Premium plan allows restaurants to be more easily discoverable, even
when a customer is not specifically searching for that particular restaurant. Customers who are enrolled in
an UberOne membership (a subscription-based program that provides discounted Uber rates and UberEats
delivery fees) do not have to pay delivery fees on their orders, which would increase restaurant demand
on the platform. In addition to tiered services based on commission rates, some platforms also provide an
auction-based advertising feature called ’Sponsored Listing.’ This allows restaurants to enhance their visi-
bility by opting to participate and setting a custom bid. They can be featured prominently when a customer
searches for a relevant keyword or browses the app. Furthermore, some platforms offer an automatic bidding
option to optimize budgets by minimizing bid amounts while maximizing the likelihood of being featured
and shown to customers.2
While the larger platforms offer more structured plans, some smaller platforms negotiate the marketing
plan and the associated commission rates with each restaurant. Figure 2 showcases the restaurant sign-up
form used by EatStreet Inc., an online food ordering platform that was founded in Madison, Wisconsin, and
is operating in over 150 markets in the US. Under the “Services” section of the sign-up form3 , the EatStreet
representative and the restaurateur agree on the percentage fees associated with the marketing plan they sign
up for.
Inspired by the current practices described above, we build the following analytical model. In our model,
there is a platform that connects restaurants to customers. Restaurants choose the commission rates they
pay, and the platform chooses a subset of restaurants to feature as well as delivery fees for each restaurant.
Both featured display and discounted delivery lead to higher restaurant demand. For each order, the platform
collects food revenue and the delivery fee from the customer, and reimburses the food revenue less the
commission to the restaurant.
Using our model, we address the following research questions. First, what are some systematic guide-
lines determining how platforms should choose delivery fees and featured restaurants, and how restaurants
should choose commission rates? Second, can we identify some problems with current practice that might
potentially explain why it is challenging for food delivery platforms to turn a profit? Third, how might we
help solve the above problems and potentially maximize the joint profits of restaurants and platform?
We find the following results. First, the platform’s optimal delivery price follows a zone-based pricing
policy, where restaurants within the “discount radius” are in the low-priced delivery zone, farther away

2
[Link] Retrieved Apr 3, 2024.
3
[Link] Retrieved September 27, 2023.
 Oh, Glaeser, and Su: Food Ordering and Delivery
4 Article submitted to Management Science

Figure 1 A tiered pricing overview by UberEats.

Figure 2 A partial snapshot of EatStreet’s restaurant sign up form.

restaurants within the “service radius” are in the high-priced delivery zone, and even farther restaurants
are not served. We find that the discount radius and the service radius increase with the commission rates
restaurants pay, which incentivizes restaurants to pay higher commissions for delivery discount. When the
platform chooses featured restaurants, it ranks the restaurants in their “virtual surplus” and grants premium
display slots to those ranked the highest. The virtual surplus of a restaurant increases with the commission
rate it chooses, and the restaurants play a bidding game to win a featured display slot. However, the virtual
surplus decreases with delivery distance since it increases the platform’s cost. When bidding for delivery
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 5

discount and/or featured display, restaurants that are very close to customers need not pay high commissions
and may still receive the benefits on the platform because of their low delivery cost. This result is reminiscent
of the internet advertising literature (e.g., Katona and Sarvary (2010) and Jerath et al. (2011)), which finds
that websites with high click-through rates need not bid and may still be ranked high and receive many
clicks. Thus, high click-through rates (for websites) and short distance from customers (for restaurants) are
similarly attractive to the platform.
We find that existing commission contracts used in the industry create inefficiency in two ways. Because
restaurants take their share of the food revenue and leave only a portion to the platform, double marginaliza-
tion results in higher delivery price and smaller market coverage than what would maximize the centralized
system’s profit. Moreover, since the restaurants do not internalize the delivery cost, restaurants situated far
away have an incentive to pay high commissions to “outbid” closer-by restaurants and win featured display
slots. In equilibrium, excess demand is directed to these distant restaurants. This outcome increases the total
cost of the system.
We show that a new contract which shares food revenue and splits delivery fees and costs in the same
fixed proportion leads the decentralized platform and the restaurant to reach the outcome that maximizes
the centralized system profit. In our numerical analysis, we calibrate our model using data on a represen-
tative city consisting of 220 market areas. For each market area, we compute the impact of our proposed
coordinating contract on platform profit, restaurant profit, and total profit. Averaged over markets, total
restaurant profit increases by 25.0% (since restaurants generally pay lower commissions and enjoy higher
demands under the coordinating contract). The platform profit increases in about one-third of the markets
but decreases in the remaining markets. Put together, the social welfare increases by 13.3% on average.
These results suggest that, the new contract can be an attractive alternative to the current food commission
contract, by alleviating restaurants’ complaint regarding high commission rates, while increasing the total
platform profit through higher demand.

2. Literature
This paper builds on the recent literature on restaurant delivery platforms by analytically modeling the
interaction between the platform and partner restaurants. Our work is most closely related to Feldman et al.
(2022) and Chen et al. (2022). They study the interaction between the platform and a single restaurant where
customers can choose between dining in and delivery. In their models, delivery orders have externalities on
dine-in customers by increasing the wait time for the shared kitchen capacity. They present revenue sharing
contracts that coordinate the system and maximize total (restaurant and platform) profit. This elegant result
no longer holds in our paper because we consider a different setting. We focus on delivery customers who do
not consider dining in, so cannibalization is not an issue. More importantly, We analyze contracts between
a platform and multiple restaurants, examining how these agreements interact or interfere with each other.
 Oh, Glaeser, and Su: Food Ordering and Delivery
6 Article submitted to Management Science

There are several papers that explore aspects of restaurant delivery other than our contracting focus. Liu
et al. (2021) focus on logistics and explore how to optimally assign customer orders to delivery drivers in
order to minimize delays. Gorbushin and Hu (2020) study courier sharing between restaurants. Li and Wang
(2020) and Cui et al. (2022) empirically study the impact of restaurant delivery platforms on restaurant
profitability. Karamshetty et al. (2020) study the impact of food delivery platforms on food waste. Mao
et al. (2022) discuss how food delivery platforms operate and how empirical research can help improve their
operational performance.
Beyond restaurant food delivery, our work also relates to the literature on online grocery delivery, which
underscores the importance of spatial aspects in online grocery operations. Belavina et al. (2017) and
Astashkina et al. (2019) study online grocery delivery, focusing on the environmental impact. They find that
smaller and more frequent grocery orders lead to less food waste and higher delivery-related emissions, but
the overall environmental impact is positive. Separately, Glaeser et al. (2019) show that optimizing pick-up
locations for online grocery retailing can have a dramatic impact on operational efficiency. More generally,
geographical considerations are also central in smart city operations. Mak (2022) describes several areas,
such as smart energy, urban logistics and autonomous vehicles.
At a more detailed level, some of our results share some parallels with the internet advertising literature,
where websites submit a bid to compete for higher positions on the search engine’s listing of sponsored
links. Edelman et al. (2007) and Varian (2007) propose the generalized second-price auction, in which
websites that bid higher per-click prices win higher ranks on search listings. Katona and Sarvary (2010) find
that the highest-ranked sites may not be the most popular sites and may not submit the highest bids, and
Jerath et al. (2011) show a “position paradox” where lower-ranked sites may receive more clicks and earn
higher revenue than higher-ranked sites. Similarly, in our analysis, we find that close-by restaurants need
not bid high commissions but may still be featured. Such display advertising strategies have been widely
studied in the literature (e.g., Liu et al. (2010), Balseiro et al. (2014), and Amaldoss et al. (2015), L’Ecuyer
et al. (2017)). We add to this literature by incorporating fulfillment costs, which is an important part of
platform’s operation, in the bidding mechanism.
Broadly, our work also adds to the literature on online marketplaces and sharing platforms. Many papers
study the implication of prices and wages on matching individual capacity with demand, particularly in the
context of ride-sharing (e.g., Banerjee et al. (2015), Cachon et al. (2017), Feng et al. (2021), Taylor (2018),
Benjaafar et al. (2021b), Besbes et al. (2021), Afeche et al. (2018), Ozkan and Ward (2017), Gurvich et al.
(2019), Bai et al. (2019), Bimpikis et al. (2019), Hu and Zhou (2019), Hu et al. (2021), Benjaafar et al.
(2021a)). Some papers study the impact of the peer-to-peer sharing economy on usage and ownership of
products (e.g., Benjaafar et al. (2018), Jiang and Tian (2016), Fraiberger and Sundararajan (2015), Filippas
et al. (2020), and Abhishek et al. (2021)). More recently, papers such as Cachon et al. (2021) and Filippas
et al. (2021) study whether prices should be set centrally or decentrally on online platforms. Our paper adds
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 7

restaurant delivery platforms, a fast-growing segment of online marketplaces and sharing platforms, to the
list of applications studied above.

3. Model
There are three parties that interact: the platform, N restaurants, and customers. Customers order food
delivery through the online platform, restaurants prepare the food, and then the platform delivers the food
to customers. Our model focuses on spatial heterogeneity: restaurants are identical in every respect other
than their location. We assume that all restaurants face unit demand, all customers are located at a single
location, and restaurant i is distance δi away. (The single customer location model makes it easier to see the
interplay of the core factors, including restaurants’ geolocational heterogeneity and misaligned incentives
among competing restaurants and the platform. In an extension, we demonstrate practical applications of
our base model across multiple market locations with varying characteristics.) Thus, the cost of fulfilling the
order from restaurant i is cδi , where c is the delivery cost per unit distance. For this order from restaurant i,
the customer pays the food price V0 (which is the same for all restaurants) plus a delivery fee pi (which may
differ across restaurants). We also normalize food cost to 0. We assume that the platform and the restaurants
maximize their profits.
The platform can influence customer demand in two ways, through its pricing and listing policies, as
described next. The platform’s pricing policy determines the delivery fees that customers pay. We assume
that the platform chooses between a low fee and a high delivery fee, i.e., pi ∈ {VL , VH }, where VL < VH .
All customers are willing to pay the former but only a fraction α (where α < 1) are willing to pay the latter.
Thus, charging a high fee for restaurant-i will decrease its demand by a factor of α. To ensure that both high
and low fees are viable options, we assume that V0 + VL > α(V0 + VH ), i.e., αVH − VL < (1 − α)V0 . In an
extension, we relax the two-point demand model and analyze linear demand model.
The second instrument available to the platform is display advertising. Specifically, the platform can
choose to feature a particular restaurant on its website more prominently. Doing so will increase the restau-
rant’s demand by a factor of β (where β > 1). However, the platform has limited space for featured display
and thus can feature at most K out of the N restaurants. Based on the empirical evidence that the effect of
sponsored advertising on conversion rates diminishes with the advertising position (Ghose and Yang 2009,
Agarwal et al. 2011), we assume that restaurants receive the same boost in revenue up to the K th slot and
beyond the K th slot, restaurants no longer benefit from the “sponsored listing.” In reality, the degree of
sales enhancement may vary among the K most sponsored restaurants, and some non-top K restaurants
might also experience advantages, depending on the design of the mobile application or webpage. We make
this simplifying assumption to highlight the impact of spatial heterogeneity in restaurants.
Consistent with practice, we study a commission contract that the platform uses to share food revenue
with the restaurants. Specifically, for each order with restaurant i, after collecting the food revenue V0 from
 Oh, Glaeser, and Su: Food Ordering and Delivery
8 Article submitted to Management Science

Restaurant-i chooses ϕi πid,p and πid,r realize

time

Platform chooses pi and S

Figure 3 Sequence of events

customers, the platform retains a fraction ϕi and sends the remaining back to the restaurant. We call ϕi the
commission rate, which is chosen by the restaurant. Higher commission rates might motivate the platform
to feature the restaurant (which increases demand by β ) or to offer discounted delivery (which increases
demand by 1/α). When negotiating with the platform, each restaurant is free to choose any commission rate
above a base minimum ϕ0 . The base commission rate ϕ0 is a fixed exogenous constant. This is the lowest
possible commission that a restaurant must pay to partner with the platform (e.g., commission payable in
the lowest pricing tier in Figure 1). At this base commission ϕ0 , we assume that it is profitable for the
platform to contract with all N restaurants, i.e., ϕ0 V0 + VH > c maxi δi ; equivalently, we drop unprofitable
restaurants from the model.
For subsequent reference, we write down the profit functions for the platform and the restaurants. The
profit of the platform associated with restaurant-i can be summarized as


β(ϕi V0 + VL − cδi ), if featured and pi = VL ,

βα(ϕi V0 + VH − cδi ), if featured and pi = VH ,
πid,p (pi , ϕi , δi ) =

ϕi V0 + VL − cδi , if not featured and pi = VL ,

α(ϕi V0 + VH − cδi ), if not featured and pi = VH ,


and the profit of restaurant-i is given as


β(1 − ϕi )V0 ,


if featured and pi = VL ,
βα(1 − ϕ )V , if featured and pi = VH ,
i 0
πid,r (pi , ϕi ) =

(1 − ϕ i )V0 , if not featured and pi = VL ,

α(1 − ϕi )V0 , if not featured and pi = VH ,


where the superscripts d, p and d, r represent “decentralized platform” and “decentralized restaurant,”
respectively. This is in contrast to the centralized system that we will later analyze to establish the first-best
benchmark.
The sequence of decisions in our model is as given in Figure 3. First, each restaurant chooses its commis-
sion rate ϕi ∈ [ϕ0 , 1] to maximize its profit πid,r (pi , ϕi ). Next, given these commission rates, the platform
sets its pricing policy (i.e., the delivery fees pi ∈ {VL , VH }) as well as its display policy (i.e., the set of K
PN
featured restaurants), in order to maximize its total profit i πid,p (pi , ϕi , δi ).
We will demonstrate later that the platform’s equilibrium pricing policy exhibits a particular structure.
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 9

We call these zone-based pricing policies, in which the platform sets delivery fees to a restaurant located
distance δ away and paying commission ϕ as

 VL , if δ ≤ x,
p(ϕ, δ) = VH , if δ ∈ (x, y],
 ∞, if δ > y.

In other words, restaurants within distance x away are in the low-priced delivery zone, farther restaurants
within distance y away are in the high-priced delivery zone, and even farther restaurants are not served. We
call x and y the discount radius and service radius, respectively.

4. Equilibrium Analysis
Our analysis proceeds as follows. In Section 4.1, we begin with the second stage subgame and analyze the
platform’s pricing and display policies. Then, in Section 4.2, we obtain the full equilibrium by studying
how restaurants choose their commission rates.

4.1. Platform Pricing and Display Policies

We begin by studying the second period subgame. Once the commission rates ϕi have been set, how does the
platform determine the delivery fees pi and the set of K restaurants to feature? The following proposition
characterizes the platform’s profit-maximizing decisions.

Proposition 1 (Second Stage Equilibrium) Given the commission rates ϕi ,

(i) the platform’s optimal pricing policy p∗i (ϕi , δi ) is a zone-based pricing policy, where the discount
radius x and the service radius y are given by

V0 VL − αVH
x(ϕi ) = ϕi · + ,
c (1 − α)c
V0 VH
y(ϕi ) = ϕi · + ,
c c

(ii) the platform’s optimal display policy is to feature the K restaurants that bring the platform the highest
profits under p∗i (ϕi , δi ) calculated above.

Proposition 1(i) shows that the optimal pricing policy in the second period subgame is a zone-based
pricing policy. This policy is applied to each restaurant separately. The platform charges the low delivery
fee VL if δi < x(ϕi ) and the high delivery fee VH if x(ϕi ) < δi < y(ϕi ). Note that as a restaurant pays
higher commission ϕi , both the discount radius x(ϕi ) and service radius y(ϕi ) correspondingly increase.
In practice, several platforms indeed offer lower delivery fees and wider service areas to restaurants paying
higher commissions.
Next, Proposition 1(ii) states that the platform’s optimal display policy is to feature the restaurants that
bring the platform the highest profits, i.e., those that maximize the value of πid,p (p∗i (ϕi , δi ), ϕi , δi ). These
 Oh, Glaeser, and Su: Food Ordering and Delivery
10 Article submitted to Management Science

platform profit shares are calculated using the optimal delivery fees p∗i (ϕi , δi ) from the zone-based pricing
policy obtained in first part of the proposition. Notice that those platform profit shares increase with com-
mission rate ϕi but decrease with delivery distance δi . In other words, the platform finds it more attractive
to feature restaurants paying higher commission ϕi but less attractive to feature a restaurant located farther
away.

4.2. Restaurants’ Choice of Commission Rates

Now, we carry on the backward induction procedure and turn to the first period. How does each restaurant
decide whether to pay the base commission ϕ0 or offer a higher commission?
1
We begin with some notation. Let Φ(m, ϕ) = 1 − m
(1 − ϕ). To interpret this expression, observe that
m(1 − Φ(m, ϕ0 )) = 1 − ϕ0 . In other words, the restaurant is indifferent between paying commission
rate Φ(m, ϕ0 ) with demand multiplied by m and paying the base commission rate ϕ0 but with demand
unchanged. This notation is useful in describing upper bounds for commissions that restaurants are willing
to pay. For example, restaurants are willing to pay commission rates of at most Φ(β, ϕ0 ) for a featured
display slot (which boosts demand by a factor β > 1). Similarly, restaurants are willing to pay at most
commission Φ( α1 , ϕ0 ) to induce the platform to charge lower delivery fees (which boosts demand by factor
1
α
). For convenience, we denote ϕα = Φ( α1 , ϕ0 ), ϕβ = Φ(β, ϕ0 ), and analogously ϕαβ = Φ(β/α, ϕ0 ). We
also introduce shorthand notations for the corresponding discount radii as x0 = x(ϕ0 ), xα = x(ϕα ), and
xαβ = x(ϕαβ ). Note that ϕ0 ≤ ϕα ≤ ϕαβ , ϕ0 ≤ ϕβ ≤ ϕαβ , and x0 ≤ xα ≤ xαβ .
Our analysis of the commission rates that restaurants are willing to pay in the first stage of the game
proceeds with the following thought exercise. We consider two separate scenarios. First, when there is abso-
lutely no chance of being featured, what is the minimum commission ϕmin the restaurant would pay, and in
particular, could it exceed the base commission ϕ0 ? Second, if the restaurant can bid a certain commission
to be guaranteed of a premium display slot, what is the maximum commission ϕmax it would pay? The
minimum and maximum commissions are characterized below, in two separate lemmas.

L EMMA 1. When there is no chance of being featured, each restaurant i chooses the commission rate

 ϕ0 , if δi ≤ x0 ,
ϕmin (δi ) = ψ(δi ), if δi ∈ (x0 , xα ],
ϕ , if δi > xα ,
0

L −αVH
where ψ(δ) = x−1 (δ) = c
V0
δ − V(1−α)V0
.

Lemma 1 considers the first scenario in our thought exercise. Interestingly, when restaurants have no
chance of being featured, they might still have an incentive to pay commissions above the required ϕ0 .
Doing so might lead the platform to lower delivery fees for the restaurant. Thus, we can interpret ϕmin as
the commission restaurants are willing to pay to secure discounted delivery. Restaurants are divided into
three zones depending on their distances from the customers. Within the base-commission discount radius
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 11

x0 , restaurants pay the base commission because doing so is sufficient for the platform to charge a low
delivery fee. Similarly, restaurants outside of distance xα also pay the base commission (but the platform
sets a high delivery fee) because they are too far away; for these restaurants, bidding up to the upper bound
of ϕα (which leaves them the same profit as simply paying ϕ0 ) still does not lead to discounted delivery.
Finally, restaurants in the intermediate zone pay the commission rate ψ(δi ); this is the minimum commis-
sion rate that induces the platform to charge a low delivery fee to a restaurant at distance δi . Thus, when
restaurants “bid” for delivery discount, restaurants within xα from customers receive discounted delivery,
i.e., endogenous bidding pushes the discount radius from x0 to xα . Note that ψ(δi ) increases in δi so that
restaurants with higher delivery distances need to pay higher commission rates in order for their customers
to receive a delivery discount. The minimum commission rate ϕmin as a function of δi is depicted in blue in
Figure 4 below.

L EMMA 2. In order to secure a premium display slot, each restaurant at distance δi chooses a commis-
sion rate that does not exceed

Φ(β, ϕ ) ϕβ , if δi ≤ x0 ,

 min 

 1 − 1 (1 − ψ(δ )), if δi ∈ (x0 , xα ],
Φ(β, ϕmin )

β i
ϕmax (δi ) = =

 Φ(β/α, ϕ min ) ϕ
 αβ
 , if δi ∈ (xα , xαβ ],
Φ(β, ϕmin ) 
ϕβ , if δi > xαβ .
The function ϕmax in Lemma 2 can be interpreted as the maximum commission restaurants are willing to
pay for premium display. Most of the restaurants are willing to pay up to Φ(β, ϕmin ) to be featured, which
gives them the same payoff as not being featured while paying the minimum commission rate ϕmin , i.e.,
β(1 − Φ(β, ϕmin )) = (1 − ϕmin ). However, restaurants at a distance between xα and xαβ are an exception.
These restaurants do not find it profitable to receive delivery discount when there is no featured display, so
ϕmin (δi ) = ϕ0 (as shown in Lemma 1). However, when featured display becomes available, these restau-
rants may consider bidding a little extra to receive delivery discount on top of featured display. Thus, the
maximum commission rate these restaurants are willing to pay is what the restaurants would pay for the
total demand increase from both premium display and discounted delivery, i.e., ϕmax = Φ( αβ , ϕmin ) = ϕαβ .
Consequently, when restaurants pay the maximum commission rate, restaurants up to xαβ away from the
customers receive delivery discount, i.e., premium display further pushes the discount radius from xα to
xαβ . Figure 4 shows the maximum commission rate ϕmax as a function of δi in red. Notice that the maxi-
mum bids ϕmax (δi ) increase over δi and then jump back down to ϕβ .
In Figure 4, the gap between the two lines represents the difference between a restaurant’s maximum
and minimum commission bids. This gap is largest for restaurants located between xα and xαβ away, due
to discontinuities in ϕmin (δi ) and ϕmax (δi ) at δi = xα and δi = xαβ . Therefore, by exploiting restaurants’
competition for scarce premium display slots, the platform can expect to receive the greatest increase in
commissions from restaurants located at distances between xα and xαβ away. Equivalently, these are also
the restaurants that might bid most aggressively for those premium display slots.
 Oh, Glaeser, and Su: Food Ordering and Delivery
12 Article submitted to Management Science

ϕαβ
—ϕ
— ϕmax
min
ψ(xα )
ϕβ

ϕ0

x0 xα xαβ δ
Figure 4 Comparison between ϕmax and ϕmin

Bearing the above interpretation in mind, we now introduce some notation. Let the virtual surplus func-
tion ∆f (δi ) be defined as ∆f (δi ) = βfmax (δi ) − fmin (δi ), where
(
ϕmin (δi )V0 + p∗i (ϕmin (δi ), δi ) − cδi , if p∗i (ϕmin (δi ), δi ) = VL ,
fmin (δi ) =
α {ϕmin (δi )V0 + p∗i (ϕmin (δi ), δi ) − cδi } , if p∗i (ϕmin (δi ), δi ) = VH
(
ϕmax (δi )V0 + p∗i (ϕmax (δi ), δi ) − cδi , if p∗i (ϕmax (δi ), δi ) = VL ,
fmax (δi ) =
α {ϕmax (δi )V0 + p∗i (ϕmax (δi ), δi ) − cδi } , if p∗i (ϕmax (δi ), δi ) = VH
Note that fmin (δi ) is the platform’s profit from restaurant i if it chooses the minimum commission, while
βfmax (δi ) is the corresponding platform profit if the restaurant chooses the maximum commission and
consequently receives a premium display slot. The difference can hence be viewed as the profit potential of
allocating a premium display slot to the restaurant. We term this the virtual surplus.
We are now ready to solve for the full equilibrium. The following proposition characterizes the commis-
sion rates that each restaurant chooses in equilibrium.

Proposition 2 (First Stage Equilibrium) In equilibrium,

(i) the K restaurants with the highest virtual surplus ∆f (δi ) choose the commission rates ϕ∗i with ϕ∗i ≥
ϕmin (δi ) such that they match the virtual surplus of the (K + 1)-th restaurant. That is,

ϕ∗i = max{ϕmin (δi ), ϕ̂i },

where ϕ̂i is defined as

(
β(ϕ̂i V0 + p∗i (ϕ̂i , δi ) − cδi ) − fmin (δi ) = ∆f (δK+1 ), if p∗i (ϕ̂i , δi ) = VL ,
βα(ϕ̂i V0 + pi (ϕ̂i , δi ) − cδi ) − fmin (δi ) = ∆f (δK+1 ), if p∗i (ϕ̂i , δi ) = VH .
∗

(ii) all other restaurants choose commission rate ϕmin (δi ).

The results of Proposition 2 can be explained by considering the profit motives of all parties involved.
For the platform, profit is maximized by allocating the premium display slots to the restaurants with the
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 13

highest virtual surplus ∆f . For these top K restaurants (in terms of virtual surplus), it is sufficient to pay
commission ϕ̂i that is just high enough to keep the marginal (K + 1)-th restaurant out of contention. Note
that ϕ̂i may be very small for some of these top K restaurants that are located close by, and the platform
may find it profitable to feature the restaurant even at its minimum commission rate ϕmin (δi ), which may
also be the same as the base commission ϕ0 . With these commissions ϕ∗i , the K restaurants with the highest
virtual surplus will indeed secure the premium display slots. Finally, since the remaining restaurants will
not receive premium display, they will simply choose the minimum commission ϕmin (δi ).

5. Centralized System
We now consider the first-best benchmark. In this scenario, the platform and the restaurants are integrated
to form a centralized system. The centralized system’s profit from restaurant-i is given as


 β(V0 + VL − cδi ), if featured and pi = VL ,

(V + V − cδ ),
0 L i if not featured and pi = VL ,
πic (pi , δi ) =

βα(V0 + VH − cδi ), if featured and pi = VH ,

α(V0 + VH − cδi ), if not featured and pi = VH ,


where the superscript c represents the centralized system. The centralized system chooses the delivery price
PN
pFi B and the set of K restaurants to feature to maximize the total profit i=1 πic (pi , δi ). The following
proposition characterizes the first-best pricing and display policies.

Proposition 3 (First Best) For a centralized platform,

(i) the first-best pricing policy pFi B (δi ) is a zone-based pricing policy, where the discount radius xF B and
the service radius y F B are given by
V0 VL − αVH
xF B = + ,
c (1 − α)c
V0 VH
yF B = + .
c c
(ii) the first-best display policy is to feature the K closest restaurants.

Let us compare the first-best policies in Proposition 3 to the equilibrium outcomes in Proposition 1. First,
the discount radius and the service radius of the first-best pricing policy are larger than those chosen in
equilibrium. That is, xF B > x(ϕi ) and y F B > y(ϕi ) for all i, as long as ϕi < 1. In other words, as long
as restaurant i retains a positive share of its food revenue, equilibrium discount and service radii will be
too small relative to first-best. The first-best and equilibrium pricing policies coincide only when ϕi = 1,
i.e., under centralization. This follows from the classic double marginalization problem in decentralized
systems. In a decentralized system, the platform receives only a fraction of the food revenue but it bears the
entire burden of delivery cost. This distorted trade-off leads to discount/service radii that are too low, which
result in higher delivery fees and smaller service coverage for the consumers.
 Oh, Glaeser, and Su: Food Ordering and Delivery
14 Article submitted to Management Science

Based on the first-best prices above, we can calculate the profit that each restaurant i can earn for the
system. Denoting the value of restaurant-i as f F B (δi ), we have

V0 + VL − cδi ,
 if δi ≤ xF B ,
FB
f (δi ) = α(V0 + VH − cδi ), if δi ∈ (xF B , y F B ],

0, otherwise.

Since f F B (δi ) is monotone decreasing in the delivery distance δi , the K closest restaurants have the high-
est f F B (δi ) values. Consequently, the centralized firm features these K restaurants that are closest to the
consumers, as shown in Proposition 3.
Although we find that the premium display slots should be allocated to the K closest restaurants in first-
best, they are allocated to the restaurants with the highest virtual surplus ∆f (δi ) in equilibrium. We stress
that restaurants with high virtual surplus may be located far away. For instance, Figure 4 suggests that those
located at distances between xα and xαβ away might have high virtual surplus. In a decentralized setting,
restaurants maximizing their own profits do not consider the platform’s delivery costs when choosing their
commission rates. As a result, restaurants with high virtual surplus but located far away may pay a very high
commission rate, thus “outbidding” restaurants closer by. Consequently, the platform allocates premium
display slots to (and hence directs demand to) these restaurants, even though it is more efficient for the
centralized system to feature closer-by restaurants with lower cost of delivery.
From the previous discussion, we see that compared to the centralized outcome, the equilibrium is inef-
ficient in two ways. In a decentralized system, restaurants take their share of the food revenue leaving only
a portion to the platform. This double marginalization makes the platform choose a smaller discount radius
and a service radius, increasing the price of delivery for customers and decreasing market coverage. In
addition, in the decentralized system, only the platform bears the cost of delivery and the restaurants do not
internalize the delivery cost when they choose the commission rates. This creates an incentive for farther-
away restaurants to outbid the closer-by restaurants and win featured display slots, which increases the total
cost of the system. In summary, the food delivery system suffers profit losses as a result of two problems:
double marginalization and inefficient bidding.

6. Food Revenue Sharing and Delivery Cost Splitting

In this section, we consider a new contract under which the decentralized platform and the restaurants can
reach the outcome of the centralized system. Previously, we have considered a commission contract in
which only the food revenue V0 is shared, while the delivery cost (and fee) pi − cδi is completely borne by
the platform; that is the industry norm. In the new contract, the platform and restaurants share the entire
profit V0 + pi − cδi . In other words, they not only share the food revenue, they also split the delivery cost
(and delivery fee).
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 15

Specifically, given any commission rate ϕi ≥ ϕ0 restaurant-i chooses, the platform pays


 β(1 − ϕi )(V0 + VL − cδi ), if featured and pi = VL ,

(1 − ϕ )(V + V − cδ ),
i 0 L i if not featured and pi = VL ,
πic,r (pi , ϕi , δi ) =

 β(1 − ϕi )α(V0 + VH − cδi ), if featured and pi = VH ,

(1 − ϕi )α(V0 + VH − cδi ), if not featured and pi = VH ,


to restaurant-i where the superscript c, r represents “coordinated restaurant.” We also call this contract the
“coordinating contract.” The following result shows that this new arrangement (of sharing both food revenue
and delivery cost) allows the system to achieve the first best outcome.

Proposition 4 With food revenue sharing and delivery cost splitting, the followings hold in equilibrium.
(i) (First Stage Equilibrium) Restaurant i chooses the commission rate ϕci given by
(   FB
ϕ0 1 f
max{ϕ0 , β + 1 − β fK+1 F B }, if δi < δK+1 ,
ϕci = i
ϕ0 , otherwise,

where fiF B = f F B (δi ) is the value of restaurant-i on the platform, as defined in Section 5.
(ii) (Second Stage Equilibrium ) The platform’s optimal pricing and display policies coincide with the
first-best policies.

Proposition 4 describes the equilibrium outcomes in the coordinating contract. The K closest restaurants
may pay a commission higher than ϕ0 , but all other restaurants pay only the base commission rate ϕ0 . A
higher commission rate ensures that the (K + 1)-th marginal restaurant remains excluded from the bid-
ding competition to be featured by the platform. For restaurants in close proximity to customers, the base
commission rate, ϕ0 , might suffice for securing a premium display slot. Otherwise, restaurants farther away
need to pay commission rate ϕi that satisfies βϕi fiF B − ϕ0 fiF B = βϕβ fK+1
FB FB
− ϕ0 fK+1 , which ensures that
it is profitable for the platform to feature restaurant i instead of the marginal restaurant.
The proposed contract aligns the platform’s and restaurants’ incentives by sharing food revenue and split-
ting delivery costs at a fixed proportion ϕi . Both parties thus receive a proportionate share of system profits,
regardless of pricing or display policies. This concept aligns with the supply chain literature (e.g., Cachon
(2003); Cachon and Lariviere (2005)), where coordinating contracts allocate a fixed percentage of total
profits to the newsvendor decision-makers by balancing underage and overage costs. In food delivery, the
contract balances “long-haul” and “short-haul” delivery economics, guiding the platform to choose optimal
discount and service radii while eliminating the incentive for distant restaurants to overpay commissions.
Note that the two inefficiencies observed in our setting have different causes. Double marginalization
arises from the separate pricing strategies employed by the platform and the restaurants as is the case in tra-
ditional supply chain contracting literature. Bidding inefficiency, on the other hand, stems from restaurants
having only a single instrument (commission rate) to steer the platform’s dual decisions (premium display
 Oh, Glaeser, and Su: Food Ordering and Delivery
16 Article submitted to Management Science

t(δi )
t(δi ) t(δi )
+ xF B yF B
δi
+ + yF B −
+ + − xF B δi −
yF B −
xF B − δi
α(VH −VL ) α(VH −VL ) VH −VL VH −VL
(a) V0 ≤ 1−α
(b) 1−α
< V0 < 1−α
(c) 1−α
≤ V0

Figure 5 Transfer t(δi ) as a function of the delivery distance δi

and delivery discount), while competing for favorable positions on the platform’s search page. Interestingly,
profit sharing emerges as a single solution for both inefficiencies.
However, the coordinating contract has limitations, as it allows only a fixed profit allocation for system
coordination: each restaurant receives a fixed fraction ϕci of total profits based on the equilibrium rate from
Proposition 4. In the supply chain literature, coordinating contracts allow greater flexibility in profit divi-
sion. To provide a similar degree of flexibility in our food delivery context, it is possible to fine-tune the
base commission rate. This is currently a free exogenous parameter, beyond the scope of our model analy-
sis, but it determines how the platform shares profits with a large number of restaurants, i.e., all those that
are not featured. Since coordination typically places most restaurants in a better position (through lower
commission costs and larger demand), the platform may contemplate increasing the base commission rate
for coordinated contracts to capture some of the resulting surplus. This potential adjustment can be quanti-
tatively evaluated by comparing the profit shares between the restaurants and the platform.
In terms of implementation, how does our coordinating contract differ from standard food commission
contracts? The key difference is captured by the term pi − cδi , which represents the delivery cost and fee
attributed to restaurant i in the coordinating contract. This term is important because in addition to sharing
a proportion of the food revenue V0 , the platform also transfers the same proportion of pi − cδi back to
restaurant-i. Let us denote the transfer amount by
(
(VL − cδi ), if pi = VL ,
t(δi ) =
α(VH − cδi ), if pi = VH .

The transfer t(δi ) can either be positive or negative. When t(δi ) is positive, the restaurant receives an addi-
tional payment over and above the food revenue (less commission); we can view this as a subsidy. When
t(δi ) is negative, the restaurant receives less than its share of food revenue; it is essentially paying a pre-
mium to reimburse the platform for high delivery costs incurred on that particular delivery. This occurs
when delivery distances are large: when delivery costs borne by the platform are high, the restaurants have
to split the delivery cost with the platform.
Observe that the transfer amount t(δi ) is a discontinuous piecewise linear function that is non-monotone
in the delivery distance. The transfer t(δi ) decreases in δi within the discount radius xF B . Then it jumps
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 17

upward at δi = xF B as the platform starts charging the high delivery fee outside the discount radius. The
profit then decreases in the delivery distance again between xF B and the service radius y F B .
The sign of the transfer amount t(δi ) also depends on the parameter values. First, in Figure 5a, when the
food revenue V0 is relatively small, the system does not have much leeway to take big losses from delivery.
As delivery distance increases, the platform increases the delivery fee quickly, using short discount/service
radii. The platform covers delivery cost with delivery fees, and shares delivery profits with the restaurants.
Next, as shown in Figure 5b, for intermediate values of food revenue V0 , regions where restaurants reimburse
the platform’s delivery costs (i.e., when t(δi ) is negative) are interspersed with regions where deliveries are
profitable (i.e., when t(δi ) is positive). In this case, the platform delivers to moderately far away customers at
a loss, but when the loss becomes too high, the platform increases the delivery price, and delivery becomes
profitable again until the delivery distance approaches the service radius. Finally, in Figure 5c, for large
values of V0 , the platform sets the discount radius and the service radius very high to induce large sales
volume. Since the food revenue V0 is high, the platform is willing to deliver for remote restaurants over
large distances even at a loss. This is so because restaurants are splitting the delivery cost with the platform.

7. Extensions
7.1. Linear Demand for Delivery
In this section, we move from a discrete menu of delivery fees, p ∈ {VH , VL }, to a continuous range of
delivery fees p ∈ [0, ∞). To do so, we first enhance our demand model by introducing customer utilities.
Specifically, each customer has a valuation V for delivery service and receives utility V − p from paying
price p for delivery. We assume that V follows a uniform distribution U [0, a]. In other words, customer
pi +
demand for restaurant-i given the delivery price pi is (1 − a
) , since market size is normalized to 1. The
profit of the platform associated with restaurant-i can be summarized as
(
+
l,p β 1 − pai (ϕi V0 + pi − cδi ), if featured,
πi (pi , ϕi , δi ) =
+
1 − pai (ϕi V0 + pi − cδi ), if not,
and the profit of restaurant-i is given as
(
+
l,r β 1 − pai (1 − ϕi )V0 , if featured,
πi (pi , ϕi , δi ) =
+
1 − pai (1 − ϕi )V0 , if not,
where the superscript l represents “linear demand model.” The decisions the platform and the restaurants
make and their sequence are the same as in the base model shown in Figure 3.
The analyses to find the equilibrium outcome with linear demand model follow the same steps as those for
the base model. Beginning with the second stage equilibrium, we find that the delivery price for restaurant-i
given the commission rate ϕi is given as

0,
 if δi ≤ ϕi Vc0 −a
pl∗
i (ϕi , δi ) =
a−ϕi V0 +cδi
2
, if δi ∈ [ ϕi Vc0 −a , ϕi Vc0 +a ],
if δi ≥ ϕi Vc0 +a .

a,
 Oh, Glaeser, and Su: Food Ordering and Delivery
18 Article submitted to Management Science

(b) (1 − ϕ0 )V0 − 2a ≤ 0 and (c) (1 − ϕ0 )V0 − 2a ≤ 0 and

{2a−(1−ϕ0 )V0 }2 {2a−(1−ϕ0 )V0 }2
(a) (1 − ϕ0 )V0 − 2a ≥ 0 β ≥ 1 + 4a{(1−ϕ )V −a}
β ≤ 1 + 4a{(1−ϕ )V −a}
0 0 0 0

Figure 6 ϕlmin (δi ) and ϕlmax (δi )

ϕi V0 −a
This delivery price has an intuitive interpretation. There is a “free delivery radius” c
within which (i.e.,
ϕi V0 −a
for restaurants δi ≤ c
) the platform charges zero delivery fee. Beyond that, there is a fixed surcharge
of c/2 for every additional unit of distance. As the price of delivery increases with distance, it eventually
exceeds the maximum willingness-to-pay a and demand falls to zero. This is analogous to the service radius
ϕi V0 −a ϕi V0 +a
in our base model. Note that both the free delivery radius c
and service radius c
increase in the
restaurant’s commission, and the delivery fee decreases and service coverage widens with the commission
ϕi . This delivery fee structure is the same as what we found in our base model.
In addition, observe that the fixed surcharge of c/2 for every additional unit of distance suggests that
the platform passes on 50% of the delivery cost c to customers. This is an artifact of the linear demand
assumption. When alternative demand functions are used, the optimal allocation of delivery cost might vary.
This is an interesting question that we leave for future research. Moreover, since our main interest lies in
understanding the structure of coordinating contracts (rather than the precise calculation of optimal prices),
we choose to use the two-type model as our base model.
Next, we proceed to analyze the first stage equilibrium in the linear demand model. To find the restau-
rants’ choice of commission rates, we need to find the commission restaurants would pay when there is no
chance of being featured ϕlmin (δi ) and the maximum commission rate restaurants are willing to pay to secure
a premium display slot ϕlmax (δi ). As was the case in the base model, restaurants are indifferent between
paying ϕlmin (δi ) without premium display and paying ϕlmax (δi ) with premium display.
With linear demand model, the shape of ϕlmin (δi ) and ϕlmax (δi ) depend on the relationship among the
parameter values, and the different cases are shown in Figure 6. The full analyses can be found in Appendix
B. Given the upper bound and the lower bound of the commissions restaurants are willing to pay, the virtual
surplus ∆f l (δi ) can be expressed as

∆f l (δi ) = βfmax
l l
(δi ) − fmin (δi ),
+
pl∗ (ϕl

l (δ ),δ )
where βfmax (δi ) = β 1 − i maxa i i (ϕlmax (δi )V0 + pl∗ l
i (ϕmax (δi ), δi ) − cδi ) is the platform’s profit

from restaurant-i when it is featured after choosing the maximum commission rate ϕlmax (δi ), and fmin
l
(δi ) =
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 19

+
pl∗ l

i (ϕmin (δi ),δi )
1− a
(ϕlmin (δi )V0 + pl∗ l
i (ϕmin (δi ), δi ) − cδi ) is the platform profit when the restaurant pays

ϕlmin (δi ) without being featured. Then, K restaurants with the highest virtual surplus ∆f l (δi ) are featured,
and they choose the commission rate ϕl∗ with ϕl∗ l
i ≥ ϕmin (δi ) such that they match the virtual surplus of

the (K + 1)-th restaurant to keep it out of the competition for premium display slots. All other restaurants
choose commission rate ϕlmin (δi ).
The first-best benchmark for the case of linear demand can be derived from the centralized system’s profit
for restaurant-i given as
(
+
β 1 − pai (V0 + pi − cδi ), if featured,
πil,c (pi , δi ) =
+
1 − pai (V0 + pi − cδi ), if not featured.

The centralized optimal delivery price pl,F B (δi ) is given as


0,
 if δi ≤ V0c−a ,
pl,F B a−V0 +cδi
if δi ∈ V0c−a , V0c+a ,
 
i (δi ) = 2
,
if δi ≥ V0c+a .

a,

Notice that the first-best delivery price pl,F

i
B
(δi ) follows the same structure as the second stage equilibrium
V0 −a
pl∗
i (δi ) above. There is a free delivery radius c
, beyond which a fixed surcharge of c/2 is levied per unit
distance. While the first-best surcharge matches the equilibrium surcharge obtained above, the first-best free
V0 −a V0 +a ϕi V0 −a ϕi V0 +a
delivery radius and service radius are larger (i.e., c
and c
compared to c
and c
before,
respectively). Also, it is optimal for the centralized system to feature the K closest restaurants.
When compared to the first-best case, the equilibrium outcome under the decentralized system shows two
inefficiencies as was the case in the base model. First, double marginalization results in smaller free delivery
radius and higher delivery price for those outside of the free delivery zone. The decrease in the free delivery
radius is analogous to the decrease in the discount radius in the base model; both are brought about by
double marginalization. Second, the restaurants that are outside of the free delivery zone have an incentive
to stretch their bids further in the premium display bidding competition to not only win featured display but
also to earn free delivery. This results in farther-away restaurants outbidding the closer-by restaurants and
winning featured display slots. From Figure 6, there are regions where the gap between the upper bound
ϕlmax (δi ) and the lower bound ϕlmin (δi ) of the restaurant commission rates widen as the delivery distance
increases ((iii) for Figure 6a; (ii) for Figure 6b; (ii) and (iii) for Figure 6c) where the restaurants with
incentives to overbid can potentially be located.
Finally, it can be shown that food revenue sharing and delivery cost splitting contract can coordinate the
system and attain the first best outcome under the linear demand model. With our proposed contract, the
twin inefficiencies of double marginalization and promoting long-distance deliveries are both eliminated.
 Oh, Glaeser, and Su: Food Ordering and Delivery
20 Article submitted to Management Science

7.2. Practical Implementation

In this section, we consider how to extend our model for practical implementation by a food delivery plat-
form. In a typical service area, the platform serves geographically dispersed demand, so we extend our
previous single-market model to incorporate multiple market locations. Each market j = 1, 2, · · · , M has
a market size mj , which can be interpreted as the population of customers concentrated at that location.
In addition, the network of partner restaurants typically reflects richer dimensions of heterogeneity (e.g.,
cuisine type, popularity, capacity constraints, priority between dine-in versus take-out, and so on). Thus, for
each restaurant i = 1, 2, · · · , N , we introduce an index of order intensity hi . Ultimately, for each (i, j) pair-
ing, the amount of orders received by the platform for restaurant i from market j is hi mj , and the delivery
distance for these orders is δij .
In this environment, each restaurant chooses commission rates and the platform sets delivery fees and
allocates display slots. Current technology allows all the above decisions to be implemented at a level
that differentiates between all restaurant-market (i, j) pairings. Specifically, each restaurant can potentially
choose a separate commission rate for each market, and the platform can potentially differentiate display
slots and delivery fees based on where the order originated. To analyze this scenario, we simply repeat
the steps of our previous analysis market-by-market. For each market j , our equilibrium analysis yields
a separate set of commission rates ϕij , delivery fees pij and featured restaurants. While technologically
feasible, this process can be quite cumbersome because restaurants would have to bid separate commission
rates for different markets.
Alternatively, in common practice each restaurant i chooses a single commission rate ϕi that is applied
to orders from all markets j . Given this commission rate, the platform then charges delivery fees and makes
featured display decisions for all restaurant-market (i, j) pairings. In general, these platform decisions may
be unconstrained (e.g., the platform can freely set pij based on delivery distance δij , and can feature dif-
ferent restaurants to different markets) or constrained (e.g., for restaurants paying high commissions, the
platform might be contractually obligated to offer discounted delivery or featured display for all markets).
As we summarize below, with additional details in Appendix C, our modeling framework can be adapted to
computationally study all these scenarios.
Similar to the base model, we begin with the second stage analysis. Given commissions ϕi chosen by
the restaurants (in the first stage), the platform charges the delivery fee p∗ij (ϕi , δij ) following Proposition 1.
That is, 
 VL , if δij ≤ x(ϕi ),
p∗ij (ϕi , δij ) = VH , if δij ∈ (x(ϕi ), y(ϕi )],
 ∞, if δ > y(ϕ ),
ij i

where x(ϕi ) and y(ϕi ) are given as in Proposition 1. With these delivery fees, the platform profit for each
p p
(i, j) pairing is πij . The platform then chooses to feature the restaurants with the highest πij for each j (if
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 21
P p
featured restaurants vary market by market), or those with the highest j πij (if choosing the same featured
set for all markets).
For the first stage analysis, restaurant commissions can be determined similarly as before. Each restaurant
i has a virtual surplus function ∆Fi , computed as the difference between the platform’s profit when the
restaurant bids its minimum and maximum possible bids. The K restaurants with the highest virtual surplus
will bid the commission that offers the platform ∆FK+1 (thus keeping the (K + 1)-th restaurant out of
contention for premium display), while all remaining restaurants bid the base commission rate.
The centralized first best outcome is obtained by setting the delivery price

FB
VL , if δij ≤ x ,

FB
pij (δij ) = VH , if δij ∈ (xF B , y F B ],
∞, if δ > y F B ,

ij

where xF B and y F B are given as in Section 5; the double marginalization problem arises as before. Also, it
is optimal for the system to feature K restaurants with highest centralized profit
 
 X X 
F F B (δ⃗i ) = hi mj (V0 + VL − cδij ) + mj ′ α(0 +VH − cδij ′ ) ,
 FB ′ FB FB

j:δij <x j :δij ′ ∈(x ,y ]

where δ⃗i = [δij ]j=1,2,·,M represents restaurant-i’s distance vector to different market locations. Since restau-
rants with a high centralized profit F F B (δ⃗i ) (i.e., those that the platform should feature to attain first best)
need not necessarily match the ones that command a high virtual surplus ∆Fim (i.e., those that the platform
features to maximize profit), there can again be an inefficiency in bidding outcomes. In other words, the
platform may feature restaurants with inefficient long-distance deliveries and/or lower order intensity.
In this extension, we see that the analysis of multiple markets follows similar steps as before. One can
carefully account for the aggregation of profit functions and equilibrium outcomes across markets. In addi-
tion, restaurant heterogeneity such as kitchen capacity and popularity can be reflected in the equilibrium
strategies of the platform following similar steps as before. Nonetheless, the analytical insights are similar
and we can apply this procedure computationally, as will be done in the case study below.

8. Illustrative Example
We calibrate the performance of the coordinating contract using data collected from a food delivery plat-
form that provides in-house delivery. For a sample city, Figure 7 illustrates restaurant and market locations
represented by triangles and circles, respectively. There are 202 restaurants and 220 markets. We define a
market as a block group, which is a subdivision of a census tract and generally contains between 600 and
3000 people. We obtain block group level population estimates from the US Census 5-year American Com-
munity Survey. In the figure, higher population markets are darker shaded than lower population markets,
based on their quantiles. We use the block group population to represent the size of each market. After
 Oh, Glaeser, and Su: Food Ordering and Delivery
22 Article submitted to Management Science

Market (Block Group Population)

445 - 1041
1041 - 1294
1294 - 1678
1678 - 2167
2167 - 6616
Market Centroid
Restaurants

Figure 7 Map of Restaurants and Markets.

separately applying our model to each market, we use the market sizes as weights to calculate the aggre-
gated platform profit and social welfare across different markets within the city. We assume that the cost of
delivery and the delivery fee for the inhabitants of each block group are, on average, the same as delivering
to the centroid, consistent with our model.
In our numerical study, we use the following parameter values. Consistent with the focal platform’s
delivery fees, we set VL = $3, VH = $6. We also set the food revenue parameter V0 = $20 to match the
platform’s minimum delivery threshold, since patrons often place minimum order sizes. The platform also
uses 5 premium display slots on its website, so we set K = 5. Next, the base commission rate is ϕ0 = 10%,
which is similar to other major delivery platforms (e.g., as described in Figure 1). We also set α = 0.7
(which is within acceptable range as implied by our assumption αVH − VL < (1 − α)V0 described above)
and β = 1.5 (which is within the ranges estimated in Baye et al., 2009, and Ghose and Yang, 2009). Finally,
similar to Belavina et al. (2017), we use c = 2 for delivery cost per mile.
We will now evaluate the coordinating contract’s performance (introduced in Section 6) compared to the
benchmark contract (from Section 4). We will examine equilibrium outcomes such as commission rates,
featured restaurants, delivery fees, platform and restaurant profits, and total profits (social welfare). To
illustrate the main insights, we will focus on a single market, detailing these outcomes for the market served
by 99 restaurants, as shown in the shaded, dotted-boundary area in Figures 8 to 11.
First, Figure 8 compares restaurant commission rates under the benchmark and coordinating contracts.
Light diamonds indicate restaurants paying the minimum 10% commission, while darker diamonds repre-
sent higher rates. Comparing Figures 8a and 8b, we see that average commission rates decrease from 12.9%
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 23

Bids - Benchmark Bids - Coordinating

0.10 0.10
0.25-0.30 0.38 - 0.40
0.50 - 0.55 Featured
Featured Market Centroid
Market Centroid Market
Market

(a) Benchmark contract. (b) Coordinating contract.

Figure 8 Restaurant Commission Bids and Featured Restaurants: Most of the restaurants pay lower
commissions under the coordinating contract

Delivery Fee - Benchmark Delivery Fee - Coordinating

Low Fee Low Fee
High Fee High Fee
Market Centroid Market Centroid
Market Market

(a) Benchmark contract. (b) Coordinating contract.

Figure 9 Delivery Fees: Discount radius is larger and greater number of restaurants receive delivery
discount under the coordinating contract.

under the benchmark contract to 11.3% under the coordinating contract. In Figure 8b, only featured restau-
rants (marked with a white ‘F’) pay more than the base commission, while others have no incentive to do
so because the platform charges first-best delivery fees regardless of commission rates. Conversely, under
the benchmark contract (Figure 8a), some restaurants pay extra commission for discounted delivery, with
higher commissions correlated to customer distance, shown by the darker diamonds.
Figure 8 shows that the sets of featured restaurants differ between the benchmark and coordinating con-
 Oh, Glaeser, and Su: Food Ordering and Delivery
24 Article submitted to Management Science

Relative Change in Restaurant Share Relative Change in Platform Share

1.08 - 1.15 0.1 - 0.5
1.15 - 1.20 0.5 - 1.0
1.20 - 1.25 1.0 - 2.0
1.25 - 1.30 2.0 - 5.0
1.30 - 1.45 5.0 - 17.5
Market Centroid Market Centroid
Market Market

(a) Restaurant profit ratio. (b) Platform profit ratio.

Figure 10 Profit ratio (profit under the coordinating contract/profit under the benchmark contract).

tracts. Under the coordinating contract, the five closest restaurants are featured, while the benchmark con-
tract features farther-away restaurants that pay high commissions not only for featured display but also for
discounted delivery. This creates system inefficiency. The coordinating contract eliminates the incentive for
distant restaurants to pay high commissions, resulting in only the closest restaurants being featured. The
average distance of featured restaurants is 1.38 miles under the benchmark contract and 0.71 miles under
the coordinating contract.
Figure 9 shows which restaurants receive delivery discount under the benchmark and coordinating con-
tracts. In Figure 9a, the benchmark contract provides discounted delivery only to restaurants close to cus-
tomers and willing to pay extra commissions. In contrast, Figure 9b shows that with the coordinating con-
tract, all restaurants receive discounted delivery as coordination eliminates double marginalization. Restau-
rants up to 4 miles away receive a discount under the coordinating contract, while the benchmark contract
limits it to 2 miles. The coordinating contract also reduces the average delivery fee from $5.75 to $3.00.
For each food order, the platform and the restaurant each earns a certain profit, and Figure 10 shows the
ratio of profit earned under the coordinating contract to the profit earned under the benchmark contract. Fig-
ures 10a and 10b respectively show the profit ratios for restaurants and platform. The green circles represent
the cases where the ratio is greater than one (i.e., higher profit under the coordinating contract), while the
yellow and red triangles represent the cases where the ratio is below one. We see that, for this particular
set of parameter values, all of the restaurants in this market benefit from the coordinating contract as their
shares increase by 8 to 45%. However, platform profitability from individual restaurants vary depending on
their delivery distance. Although platform profits from individual restaurants might decrease, total platform
profit (aggregated over all restaurants) increases by 13.8%. In other words, the coordinating contract makes
all parties better off in this market.
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 25

Relative Change in Restaurant Share Relative Change in Platform Share

1.08 - 1.15 0.1 - 0.5
1.15 - 1.20 1.0 - 2.0
1.30 - 1.45 5.0 - 17.5
Restaurant 1 Restaurant 1
Restaurant 2 Restaurant 2
Restaurant 3 Restaurant 3
Market Centroid Market Centroid
Market Market

(a) Restaurant profit ratio. (b) Platform profit ratio.

Figure 11 Profit ratio for a subset of three selected restaurants.

1 2 3 1 2 3
20 15

15
Profit Per Order

Profit Per Order

10

Food Revenue Food Commission

10
Delivery Cost Delivery Cost
Delivery Revenue Delivery Revenue
5
5

0 0

Benchmark Coordinating Benchmark Coordinating Benchmark Coordinating Benchmark Coordinating Benchmark Coordinating Benchmark Coordinating
Profit Comparison for Individaul Restaurants Platform's Profit Comparison for Individaul Restaurants

(a) Restaurant profit components. (b) Platform profit components.

Figure 12 Comparison of profit components under benchmark and coordinating contracts, for three
selected restaurants.

We focus on three representative restaurants labeled Restaurants 1, 2, and 3 in Figure 11 to analyze profit
components such as food revenue, commission, delivery cost, and fee under the coordinating contract. Per-
order profit includes food revenue (V0 , set at $20 in this example) and delivery profit (pi − cδi , which can
vary). This profit can be adjusted based on delivery price (by α) and display choice (by β ), with the total gain
split between the platform and the restaurant according to commission bid ϕi . Each of these components
may change differently, as summarized in Figure 12.
First, Restaurant 1, located 0.74 miles from the market, is featured under the coordinating contract but
not under the benchmark. Due to display advertising, total platform and restaurant profits (total height of
colored bars) increase under the coordinating contract, as shown in the leftmost bars of Figures 12a and
12b. Profits for both parties rise, with delivery profit (previously kept solely by the platform) now shared
with the restaurant. The positive transfer amount, t(δi ) = (1 − ϕi )(VL − cδi ), is added to the restaurant’s
profit, represented by the tan-colored bar.
 Oh, Glaeser, and Su: Food Ordering and Delivery
26 Article submitted to Management Science

Restaurant 2, located 1.66 miles from the market, is featured under the benchmark contract (by paying
a high 50% commission) but not under the coordinating contract. With significantly reduced commissions
under the coordinating contract, platform profit drops while restaurant profit increases. Both contracts offer
discounted delivery, but the delivery loss (VL − cδi ), previously borne solely by the platform under the
benchmark contract, is shared with the restaurant in the coordinating contract. The transparent section of
the bar graphs in Figure 12 represents the delivery loss.
Restaurant 3, 3.88 miles away, is not featured under either contract and pays the base commission. It
receives discounted delivery under the coordinating contract but not under the benchmark contract. The
coordinating contract eliminates double marginalization, reducing delivery costs and scaling up the restau-
rant’s total gain by 1/α. The restaurant then shares the delivery loss (VL − cδi ) with the platform, allowing
both to enjoy increased profits under the coordinating contract.
The detailed computational results described above apply to our focal market, but we can adopt the same
procedure to obtain results for all 220 market areas in our dataset. There are two possible approaches to
carry out our analysis across markets. First, we can repeat our analysis market-by-market, and this approach
accommodates maximum flexibility, whereby all decisions (commission rates, delivery fees, premium dis-
play) are individually and independently set for each market. Alternatively, some decisions might be con-
strained to be the same across markets. Each restaurant may bid a single commission rate for all markets,
and based on restaurant bids, the platform may feature the same set of restaurants for all markets. The equi-
librium outcomes for this alternative are shown in Figure 13. We see that, similar to Figure 8, commission
rates are higher under the benchmark contract; in the coordinating contract, only featured restaurants bid
above the base commission of 10%, but in the benchmark contracts, many restaurants bid above the base
commission, presumably to receive discounted delivery. The set of featured restaurants appear to be the
most centrally-located restaurants with the lowest delivery costs. For both approaches, the qualitative nature
of equilibrium results appear similar.
With the equilibrium results for all 220 markets in hand, we can draw comparisons across markets. Our
goal is to identify market areas where our proposed coordinating contract might have the greatest impact and
provide insights on potential implementation of our proposed coordinating contract. We begin by calculating
the total profit of the platform and restaurants (which we call market aggregated social welfare) for each
market. We find that market aggregated social welfare always increases when the coordinating contract is
adopted. As shown in Figure 14a, the increase is usually larger (represented by darker shades) for markets
that are more distant from the restaurants. This increase can be as high as 23% and averages 15% over
all markets. However, as shown in Figure 14b, platform profit ratios can be greater or less than one. In
other words, the coordinating contract benefits the platform only in some markets (represented by shades of
green) that again happen to be markets situated farther away from most of the restaurants. For these markets
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 27

One Bid for Market - Benchmark Bids One Bid for Market - Coordinating Bids
0.30 - 0.50 0.10
0.50 - 0.80 0.30 - 0.50
Featured Featured
Market Market

(a) Benchmark Contract. (b) Coordinating Contract.

Figure 13 Restaurant Commission Bids and Featured Restaurants - Bid for the Whole Market: Most of the
restaurants continue to pay lower commissions under the coordinating contract.

Aggregated Social Welfare Comparison Aggregated Platform Comparison

1.04 - 1.10 0.00 - 0.60
1.10 - 1.15 0.60 - 0.75
1.15 - 1.20 0.75 - 1.00
1.20 - 1.25 1.00 - 1.50
Restaurants 1.50 - 44.10

(a) Change in Market Aggregated Social Welfare. (b) Change in Market Aggregated Platform Share.
Figure 14 Change in Market Aggregated Social Welfare and Platform Share.

(30.9% of markets and 33.1% of population), the coordinating contract results in a win-win outcome for the
platform and restaurants.
In practice, our numerical results suggest that the coordinating contract can be introduced in stages,
starting with suburban markets before moving to central ones. In suburban areas, such as the example of
Restaurant 3, the coordinating contract benefits all parties as many restaurants are relatively remote. The
 Oh, Glaeser, and Su: Food Ordering and Delivery
28 Article submitted to Management Science

(b) Relative share of platform as average distance variesa .

(a) Relative share of restaurants as average distance varies.
a: Two data points with relative change in share greater than 10 are not shown.
Figure 15 Effect of Distance on Relative Change in Restaurant and Platform Share

platform lowers delivery fees to boost demand, while restaurants agree to share delivery costs, creating a
mutually beneficial arrangement ideal for long-distance deliveries.
In more central areas, the coordinating contract benefits restaurants but negatively impacts the platform
because restaurants pay lower commission. To address this, the platform could consider raising the base
commission fee (as discussed in Section 6) while ensuring restaurant profits remain protected. This is fea-
sible since restaurants gain from higher demand (thanks to lower delivery fees and wider service coverage)
and reduced commission rates under the coordinating contract.
To offer a comprehensive overview, Figure 15 shows scatter plots illustrating the relative change in aggre-
gate shares for restaurants and the platform in each market against the average distance between restaurants
and customers. According to our current parameters, restaurant profits increase in every market (all data
points are above the y = 1 dotted line) as shown in Figure 15a. However, as noted in Figure 14b and again
in Figure 15b, the platform’s share only increases in about 31% of markets, most of which are around 3
miles away on average.

9. Conclusion
How should food ordering and delivery platforms split the pie with partner restaurants? The prevailing
industry practice, in the form of commission contracts, has evolved steadily over the past decade or so.
The earliest and simplest approach is for platforms to negotiate the commission rate with each restaurant
separately. A popular next step is to introduce an auction-like bidding process – restaurants can choose
to pay extra commissions for a chance to secure limited premium display slots on the platform’s app or
website. Most recently, commission rates have also been tied to delivery fees, i.e., restaurants paying higher
commissions enjoy discounted or even free delivery. All these innovations boost customer demand and thus
enlarge the pie, but they focus only on one part of the pie associated with food revenues. The total pie, in
fact, includes food revenues as well as delivery costs and fees. In this paper, we argue that platforms should
also split the part of the pie associated with delivery. Specifically, restaurants located far away should split
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 29

the delivery cost with platforms, while restaurants located close by should enjoy a share of the delivery
profit. Doing so aligns incentives perfectly between platforms and restaurants, and we show that it is the
only way to truly maximize the total pie.
Nonetheless, our recommendations come with two important caveats. First, although restaurants can eas-
ily see how much customers are paying for their orders, they do not know the precise cost incurred by the
platform to deliver these orders. This classic information asymmetry issue might explain why existing com-
mission contracts have focused exclusively on the food revenue side. Fortunately, there are several readily
available proxies for delivery cost, such as distance and time taken for the delivery. In fact, some platforms
are fully transparent and share the locations of all orders with partner restaurants. Moreover, aggregate data
(e.g., total delivery miles over the month), which is easier to estimate, suffices for implementation purposes.
Second, although our analysis in this paper assumes that all parties are profit-maximizing, it is possible
that some platforms might have alternative objectives. For instance, a platform new to a city might prefer
to build a customer base, possibly at a loss. In this case, long-distance deliveries might be attractive to the
platform but not to restaurants, and splitting delivery costs might be easier said than done. To safeguard
against these scenarios, restaurants and platforms need to elucidate clear geographic boundaries.
Moving ahead, in this area of contract design for restaurant delivery platforms, we feel that a useful next
step is to incorporate platform interactions. In this paper, we model a monopoly platform, but in practice
several restaurant delivery platforms co-exist. We find it worthwhile to disentangle platform demand (e.g.,
customers going to a particular platform to search for restaurants) from restaurant demand (e.g., customers
having a specific restaurant in mind before ordering from the platform). Once the demand system is spec-
ified, our model can be applied at each platform to answer interesting questions, such as: How do delivery
fees and commission rates depend on the network structure (i.e., which restaurants are on which platforms)?
Should restaurants participate in every platform or be exclusive partners on one platform? Is the existing
market structure stable or will it evolve such that platforms will eventually become local monopolies? We
leave these exciting questions for future research.

References
Abhishek V, Guajardo JA, Zhang Z (2021) Business models in the sharing economy: Manufacturing durable goods in
the presence of peer-to-peer rental markets. Information Systems Research 32(4):1450–1469.

Afeche P, Liu Z, Maglaras C (2018) Ride-hailing networks with strategic drivers: The impact of platform control
capabilities on performance. Columbia Business School Research Paper (18-19):18–19.

Agarwal A, Hosanagar K, Smith MD (2011) Location, location, location: An analysis of profitability of position in
online advertising markets. Journal of marketing research 48(6):1057–1073.

Amaldoss W, Desai PS, Shin W (2015) Keyword search advertising and first-page bid estimates: A strategic analysis.
Management Science 61(3):507–519.
 Oh, Glaeser, and Su: Food Ordering and Delivery
30 Article submitted to Management Science

Astashkina E, Belavina E, Marinesi S (2019) The environmental impact of the advent of online grocery retailing.
Available at SSRN 3358664 .

Bai J, So KC, Tang CS, Chen X, Wang H (2019) Coordinating supply and demand on an on-demand service platform
with impatient customers. Manufacturing & Service Operations Management 21(3):556–570.

Balseiro SR, Feldman J, Mirrokni V, Muthukrishnan S (2014) Yield optimization of display advertising with ad
exchange. Management Science 60(12):2886–2907.

Banerjee S, Riquelme C, Johari R (2015) Pricing in ride-share platforms: A queueing-theoretic approach. Available at
SSRN 2568258 .

Baye MR, Gatti JRJ, Kattuman P, Morgan J (2009) Clicks, discontinuities, and firm demand online. Journal of Eco-
nomics & Management Strategy 18(4):935–975.

Belavina E, Girotra K, Kabra A (2017) Online grocery retail: Revenue models and environmental impact. Management
Science 63(6):1781–1799.

Benjaafar S, Bernhard H, Courcoubetis C, Kanakakis M, Papafragkos S (2021a) Drivers, riders, and service providers:
The impact of the sharing economy on mobility. Management Science .

Benjaafar S, Ding JY, Kong G, Taylor T (2021b) Labor welfare in on-demand service platforms. Manufacturing &
Service Operations Management .

Benjaafar S, Kong G, Li X, Courcoubetis C (2018) Peer-to-peer product sharing: Implications for ownership, usage,
and social welfare in the sharing economy. Management Science 65(2):477–493.

Besbes O, Castro F, Lobel I (2021) Surge pricing and its spatial supply response. Management Science 67(3):1350–
1367.

Bimpikis K, Candogan O, Saban D (2019) Spatial pricing in ride-sharing networks. Operations Research 67(3):744–
769.

Cachon GP (2003) Supply chain coordination with contracts. Handbooks in operations research and management
science 11:227–339.

Cachon GP, Daniels KM, Lobel R (2017) The role of surge pricing on a service platform with self-scheduling capacity.
Manufacturing & Service Operations Management 19(3):368–384.

Cachon GP, Dizdarer T, Tsoukalas G (2021) Decentralized or centralized control of online service platforms: Who
should set prices? Available at SSRN .

Cachon GP, Lariviere MA (2005) Supply chain coordination with revenue-sharing contracts: strengths and limitations.
Management science 51(1):30–44.

Chen M, Hu M, Wang J (2022) Food delivery service and restaurant: Friend or foe? Management Science .

Cui H, Niu R, Tong X (2022) Food in the time of covid: Estimating the effects of delivery platforms on small-business
restaurants in california. Available at SSRN 4219542 .
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 31

Davalos J (2021) New york city council passes bill to cap food delivery fees. Bloomberg August
26, 2021, [Link]
risks-markets-wrap.

Dickinson G (2023) How to navigate and negotiate contracts with third-party food delivery platforms
April 17, 2023, [Link]
delivery-platforms-commission-rates.

Edelman B, Ostrovsky M, Schwarz M (2007) Internet advertising and the generalized second-price auction: Selling
billions of dollars worth of keywords. American economic review 97(1):242–259.

Feldman P, Frazelle AE, Swinney R (2022) Managing relationships between restaurants and food delivery platforms:
Conflict, contracts, and coordination. Management Science .

Feng G, Kong G, Wang Z (2021) We are on the way: Analysis of on-demand ride-hailing systems. Manufacturing &
Service Operations Management 23(5):1237–1256.

Filippas A, Horton JJ, Zeckhauser RJ (2020) Owning, using, and renting: Some simple economics of the “sharing
economy”. Management Science 66(9):4152–4172.

Filippas A, Jagabathula S, Sundararajan A (2021) The limits of centralized pricing in online marketplaces and the
value of user control. Technical report, Working Paper.

Forman L (2019) Some restaurants get served in delivery boom. The Wall Street Journal June 13, 2019,
[Link]

Fraiberger SP, Sundararajan A (2015) Peer-to-peer rental markets in the sharing economy. NYU Stern School of Busi-
ness research paper 6.

Ghose A, Yang S (2009) An empirical analysis of search engine advertising: Sponsored search in electronic markets.
Management science 55(10):1605–1622.

Glaeser CK, Fisher M, Su X (2019) Optimal retail location: Empirical methodology and application to practice:
Finalist–2017 m&som practice-based research competition. Manufacturing & Service Operations Management
21(1):86–102.

Gorbushin A, Hu M (2020) Courier sharing in food delivery. Available at SSRN 3548978 .

Gurvich I, Lariviere M, Moreno A (2019) Operations in the on-demand economy: Staffing services with self-
scheduling capacity. Sharing Economy, 249–278 (Springer).

Hu B, Hu M, Zhu H (2021) Surge pricing and two-sided temporal responses in ride hailing. Manufacturing & Service
Operations Management .

Hu M, Zhou Y (2019) Price, wage and fixed commission in on-demand matching. Available at SSRN 2949513 .

Jerath K, Ma L, Park YH, Srinivasan K (2011) A “position paradox” in sponsored search auctions. Marketing Science
30(4):612–627.
 Oh, Glaeser, and Su: Food Ordering and Delivery
32 Article submitted to Management Science

Jiang B, Tian L (2016) Collaborative consumption: Strategic and economic implications of product sharing. Manage-
ment Science 64(3):1171–1188.

Karamshetty V, Freeman M, Hasija S (2020) An unintended consequence of platform dependence: empirical evidence
from food-delivery platforms .

Katona Z, Sarvary M (2010) The race for sponsored links: Bidding patterns for search advertising. Marketing Science
29(2):199–215.

L’Ecuyer P, Maillé P, Stier-Moses NE, Tuffin B (2017) Revenue-maximizing rankings for online platforms with
quality-sensitive consumers. Operations Research 65(2):408–423.

Li Z, Wang G (2020) The role of on-demand delivery platforms in restaurants. Available at SSRN 3813891 .

Littman J (2019) Delivery by the numbers: How top third-party plat-

forms compare URL [Link]
delivery-by-the-numbers-how-top-third-party-platforms-compare/564279/.

Liu D, Chen J, Whinston AB (2010) Ex ante information and the design of keyword auctions. Information Systems
Research 21(1):133–153.

Liu S, He L, Shen ZJM (2021) On-time last-mile delivery: Order assignment with travel-time predictors. Management
Science 67(7):4095–4119.

Mak HY (2022) Enabling smarter cities with operations management. Manufacturing & Service Operations Manage-
ment 24(1):24–39.

Mao W, Ming L, Rong Y, Tang CS, Zheng H (2022) On-demand meal delivery platforms: Operational level data and
research opportunities. Manufacturing & Service Operations Management .

Oblander ES, McCarthy D (2021) How has covid-19 impacted customer relationship dynamics at restaurant food
delivery businesses? [Link] doi. org/10 23:2139.

Ozkan E, Ward A (2017) Dynamic matching for real-time ridesharing. Available at SSRN 2844451 .

Rana P, Haddon H (2021) Doordash and uber eats are hot. they’re still not making money. The Wall Street
Journal May 28, 2021, [Link]
money-11622194203.

Sullivan M (2023) Price controls in a multi-sided market .

Taylor TA (2018) On-demand service platforms. Manufacturing & Service Operations Management 20(4):704–720.

Varian HR (2007) Position auctions. international Journal of industrial Organization 25(6):1163–1178.

Yahoo Finance (2021) Just eat [Link] n.v. (grub) income statement URL [Link]
com/quote/GRUB/financials?p=GRUB.

Yahoo Finance (2022a) Doordash, inc. (dash) income statement URL [Link]
DASH/financials?p=DASH.
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 33

Yahoo Finance (2022b) Just eat [Link] n.v. (jtkwy) income statement URL [Link]
com/quote/JTKWY/financials?p=JTKWY.

Yahoo Finance (2022c) Uber technologies, inc. (uber) income statement URL [Link]
quote/UBER/financials?p=UBER.
 Oh, Glaeser, and Su: Food Ordering and Delivery
34 Article submitted to Management Science

Appendix A. Proofs
Proof of Proposition 1 (i): Given ϕi , platform’s optimal pricing policy p∗ (ϕi , δi ) should satisfy p∗ (ϕi , δi ) ∈
arg maxp πid,p (p, ϕi ). The optimal prices can be characterized as the following.
(a) The optimal price for the platform is VL when ϕi V0 + VL − cδi ≥ α(ϕi V0 + VH − cδi ) and ϕi V0 + VL −
L −αVH
cδi ≥ 0. The first condition holds when δi ≤ ϕi Vc0 + V(1−α)c , and the second condition holds when δi ≤
VL −αVH L −αVH
ϕi Vc0 + VcL . Since (1−α)c
< VL
c
, it is optimal for the platform to charge VL when δi ≤ ϕi Vc0 + V(1−α)c ,
that is, when δi ≤ x(ϕi )
(b) The optimal price for the platform is VH when α(ϕi V0 + VH − cδi ) > ϕi V0 + VL − cδi and α(ϕi V0 +
VL −αVH
VH − cδi ) ≥ 0. The first condition holds when δi > ϕi Vc0 + (1−α)c
, and the second condition holds
VH
when δi ≤ ϕi Vc0 + c
. Therefore, VH is the optimal price when δi ∈ (x(ϕi ), y(ϕi )].
(c) The optimal price for the platform is ∞ when 0 > ϕi V0 + VL − cδi and 0 > α(ϕi V0 + VH − cδi ). The
first condition holds when δi > ϕi Vc0 + VcL , and the second condition holds when δi > ϕi Vc0 + VcH . The
second condition implies the first. Therefore, ∞ becomes the optimal price when δi > ϕi Vc0 + VcH , that
is, when δi > y(ϕi ).
■

Proof of Proposition 1 (ii): The platform’s optimal display decision in equilibrium is to feature K
restaurants that bring the largest profit increase to the platform when featured. We provide the formal proof
of this statement before the proof of Proposition 2, which comes after the manuscript defines the necessary
notations. ■

Proof of Lemma 1: Let πid,r (ϕi , δi ) represent restaurant-i’s profit given ϕi and δi . Since there is no consid-
eration of premium display slot allocation, platform only decides whether to charge a high or a low price
for delivery. The restaurants only consider whether they want high or low price for delivery when choosing
their commission bids given the platform’s pricing strategy given in Proposition 1 (i).
(1) δi ≤ x0 : In this case, from Proposition 1 (i), the platform chooses pi = VL even when the restaurant
pays ϕi = ϕ0 . Therefore, the restaurant chooses to bid ϕmin (δi ) = ϕ0 and receives low price for delivery.
Note that ϕ0 is the profit maximizing bid since for any ϕ > ϕ0 , πid,r (ϕ, δi ) = (1 − ϕ)V0 < (1 − ϕ0 )V0 .
L −αVH
(2) δi ∈ (x0 , xα ]: Note that ψ(δ) = x−1 (δ) = c
V0
δ − V(1−α)V0
is the minimum bid that guarantees low price
delivery as x(ϕ) is the discount radius defined in Proposition 1, which determines the distance limit
under which restaurants receive delivery discounts. Given that the platform follows its best response
strategy described in Proposition 1 (i), we show that ψ(δi ) is restaurant-i’s unique best response
strategy. Consider any deviation ϕ with ϕ > ψ(δi ). Then, πr (ϕ, δi ) = (1 − ϕ)V0 < (1 − ψ(δi ))V0 =
πr (ψ(δi ), δi ). Consider deviation ϕ with ϕ < ψ(δi ). Then, πr (ϕ, δi ) = α(1 − ϕ)V0 ≤ α(1 − ϕ0 )V0 =
(1−ϕα )V0 ≤ (1−ψ(δi ))V0 = πr (ψ(δi ), δi ). The first equality follows from the fact that ϕ < ψ(δi ). The
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 35

first inequality holds since ϕ ≥ ϕ0 . The second equality holds from the definition of ϕα , and the second
inequality holds since ψ(δi ) ≤ ϕα for δi ≤ xα . Therefore, ϕi = ψ(δi ) uniquely maximizes restaurant-i’s
profit.
(3) δi > xα : In this case, there is no profitable way for a restaurant to receive delivery discount. Therefore,
the restaurant chooses to bid ϕα∗
i = ϕ0 and receives high price for delivery. Given platform’s pricing

strategy, we show that there is no profitable deviation for the restaurant. Consider deviation ϕ with
ϕ ∈ (ϕ0 , ψ(δi )). Then, πid,r (ϕ, δi ) = α(1 − ϕ)V0 < α(1 − ϕ0 )V0 = πid,r (ϕ0 , δi ). The first equality holds
because bidding less than ψ(δi ) gives high price for delivery. The first inequality follows from ϕ > ϕ0 .
Consider deviation ϕ with ϕ ≥ ψ(δi ). Then, πid,r (ϕ, δi ) = (1 − ϕ)V0 ≤ (1 − ψ(δi ))V0 < (1 − ϕα )V0 =
α(1 − ϕ0 )V0 = πid,r (ϕ0 , δi ). The first equality holds since bidding higher than ψ(δi ) guarantees low
price for delivery. The first inequality holds since ϕ ≥ ψ(δi ), and the second inequality holds because
ψ(δi ) > ϕα for δi > xα . The second equality holds from the definition of ϕα .
From (1), (2), and (3) we see that for each restaurant ϕmin (δi ) is the unique profit maximizing commission
rate. ■

Proof of Lemma 2: This lemma also means that ϕmax (δ) is the maximum commission rate a restaurant is
willing to pay for possible upgrades.
(1) δi ≤ x0 : In this case, ϕmax (δ) = ϕβ . Restaurants within x0 are charged VL for the delivery with the
base bid ϕ0 , and the restaurant only considers bidding more to qualify for a premium display slot. The
maximum bid the restaurant is willing to pay for a premium slot should satisfy β(1 − ϕmax (δ))V0 =
(1 − ϕ0 )V0 . Therefore, ϕmax (δ) = Φ(β, ϕmin ) = ϕβ .
(2) δi ∈ (x0 , xα ]: From Lemma 1, we know that it is profit maximizing for a restaurant with δi ∈ (x0 , xα ]
to bid ψ(δi ) without the consideration of premium display slots. Maximum commission the restaurant
is willing to bid for a premium display slot then should satisfy β(1 − ϕmax (δi ))V0 = (1 − ψ(δi ))V0 .
Therefore, ϕmax (δi ) = Φ(β, ψ(δi )) = 1 − β1 (1 − ψ(δi )).
(3) δi ∈ (xα , xαβ ]: From Lemma 1, we know that the minimum bid the restaurants would pay without the
consideration of premium display slot allocation is ϕ0 in this case, which gives high delivery price.
Therefore, the restaurant has an option to pay for both premium display and delivery discount. The
maximum bid the restaurant is willing to pay thus should satisfy β(1 − ϕmax (δi ))V0 = α(1 − ϕ0 )V0 .
Therefore, ϕmax (δi ) = Φ(β/α, ϕ0 ) = ϕαβ .
(4) δi > xαβ : The minimum bid the restaurants are willing to pay without the consideration of premium slot
allocation in this case is ϕ0 , which gives high price for delivery. Note that for restaurants with δi > xαβ
there is no profitable bid ϕ, which will guarantee low price of delivery. This is because the maximum
possible commission a restaurant is willing to pay for both premium display and delivery discount is
ϕαβ . However, since δi > xαβ , bidding ϕαβ will still give high price for delivery as one can see from
 Oh, Glaeser, and Su: Food Ordering and Delivery
36 Article submitted to Management Science

Proposition 1. Therefore, the restaurant would only pay for a premium display slot. The maximum bid
ϕmax (δi ) should satisfy βα(1 − ϕmax (δi ))V0 = α(1 − ϕ0 )V0 , and ϕmax (δi ) = Φ(β, ϕ0 ) = ϕβ .
■

Proof of Proposition 1 (ii): Given any commission rate ϕi from i = 1, 2, · · · , N , let κ denote the set of
restaurants the platform decides to allocate premium display slots. Then, the platform’s profit π d,p (ϕ, κ) is
given as
N
X X
d,p
π (ϕ, κ) = fmin (δi ) + (βf (ϕj , δj ) − fmin (δj )),
i=1 j∈κ

where (
ϕi V0 + VL − cδi , if pi = VL ,
f (ϕi , δi ) =
α(ϕi V0 + VH − cδi ), if pi = VH .
Therefore, the unique best response strategy for the platform is to allocate the premium display slots to K
restaurants with the highest βf (ϕj , δj ) − fmin (δj ) values. ■

From the following lemmas we show that if restaurants’ equilibrium best response strategy exists,
it has to be unique in the way that K restaurants with the highest ∆f (δi ) are featured and no other
equilibrium structure exists. For the rest of the proofs recall that the restaurants are ordered so that
∆f (δ1 ) > ∆f (δ2 ) > · · · > ∆f (δN ).

L EMMA 3. Assume that restaurant-j bids to be featured as a best response strategy. Then, it is profit-
maximizing for any restaurant with i < j to bid to be featured.

Proof of Lemma 3: If restaurant-j is featured, then from Proposition 1 (ii) it means that restaurant-j found a
commission rate ϕj that satisfies ϕj ≤ ϕmax (δj ) where βf (ϕj , δj ) − fmin (δj ) is among the K highest. Then,
for any restaurant-i with i < j can find ϕi which satisfies βf (ϕi , δi ) − fmin (δi ) > βf (ϕj , δj ) − fmin (δj ) and
ϕi < ϕmax (δi ). Since ϕi < ϕmax (δi ), it is more profitable for restaurant-i to bid ϕi and be featured than to
bid ϕmin (δi ) and not be featured. ■

L EMMA 4. Assume that i < j , but ϕi and ϕj are set so that restaurant-j is featured but restaurant-i is
not. Then, ϕi and ϕj cannot form an equilibrium best response strategy.

Proof of Lemma 4: Assume that restaurant-j is featured, but restaurant-i is not. Then their bids should sat-
isfy βf (ϕj , δj ) − fmin (δj ) > βf (ϕi , δi ) − fmin (δi ). Since restaurant-i is not featured with such commission
rate ϕi ≥ ϕmin (δi ), restaurant-i’s profit is (1 − ϕi )V0 ≤ (1 − ϕmin (δi ))V0 . Now consider a deviation ϕ̂i <
ϕmax (δi ) where βf (ϕ̂i , δi ) − fmin (δi ) = (βf (ϕj , δj ) − fmin (δj )) + ϵ < βfmax (δi ) − fmin (δi ) with ϵ > 0 being
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 37

a small positive number. Then, bidding ϕ̂i wins restaurant-i a featured display slot and doing so is more
profitable for restaurant-i since β(1 − ϕ̂i )V0 > β(1 − ϕmax (δi ))V0 = (1 − ϕmin (δi ))V0 ≥ (1 − ϕi )V0 . The
first inequality holds from ϕ̂i < ϕmax (δi ); the equality holds from β(1 − ϕmax (δi ))V0 = (1 − ϕmin (δi ))V0 ;
the last inequality holds from ϕ̂i ≥ ϕmin (δi ). Therefore, a profitable deviation ϕ̂i exists and ϕi cannot form
an equilibrium best response strategy. ■

From Lemma 3 and Lemma 4, we can conclude that if restaurants’ equilibrium best response strategy
exists, it has to be unique in the way that K restaurants with the highest ∆f (δi ) are featured and no other
equilibrium structure exists.

Proof of Proposition 2: Given platform allocates the premium display slots to the K restaurants with the
highest βf (ϕj , δj ) − fmin (δj ) values as shown in Proposition 1 (ii), we show that ϕ∗i is restaurant-i’s unique
best response strategy.
For restaurants with i < K + 1, the unique best response strategy is to choose the ϕi that brings the
highest restaurant profit that satisfies βf (ϕi , δi ) − fmin (δi ) ≥ (βf (ϕmax (δK+1 ), δK+1 ) − fmin (δK+1 )). For
simplicity of notation, we use ∆f ∗ ≡ ∆f (δK+1 ). We show this by contradiction. If restaurant-i with i <
K + 1 chooses ϕi with βf (ϕi , δi ) − fmin (δi ) < (βf (ϕmax (δK+1 ), δK+1 ) − fmin (δK+1 )), then restaurant K +
1 can win a featured display slot by bidding ϕK+1 which satisfies βf (ϕi , δi )−fmin (δi ) < βf (ϕK+1 , δK+1 )−
fmin (δK+1 ) < βf (ϕmax (δK+1 ), δK+1 ) − fmin (δK+1 ). Then, restaurant-i loses the featured display slot and
earns (1 − ϕi )V0 ≤ (1 − ϕmin (δi ))V0 = β(1 − ϕmax (δi ))V0 < β(1 − ϕ∗i )V0 . Therefore, restaurant-i has an
incentive to deviate to ϕi with βf (ϕi , δi ) − fmin (δi ) ≥ (βf (ϕmax (δK+1 ), δK+1 ) − fmin (δK+1 )). Therefore,
ϕi with βf (ϕi , δi ) − fmin (δi ) < (βf (ϕmax (δK+1 ), δK+1 ) − fmin (δK+1 )) cannot form an equilibrium best
response strategy.
Below, we show that the equilibrium does exist by showing that there is no incentive for any of the
restaurants to unilaterally deviate from bidding ϕ∗i given that the platform allocates premium display slots
to i ∈ {1, 2, · · · , K}.
(1) δi ≤ x0 : The equilibrium bid ϕ∗i for restaurant-i can be characterized as the following.

∗
ϕi = ϕ0 ,
 if βfmin (δi ) > fmin (δi ) + ∆f ∗ ,
βf (ϕ∗i , δi ) = fmin (δi ) + ∆f ∗ , if βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βfmin (δi ),
ϕ∗ = ϕ ,

otherwise.
i 0

(a) If βfmin (δi ) > fmin (δi ) + ∆f ∗ :

We show that there is no incentive for the restaurant to unilaterally deviate from ϕ∗i = ϕ0 . In this
case, i ∈ {1, 2, · · · , K}, and bidding ϕ∗i = ϕ0 will give πid,r (ϕ∗i , δi ) = β(1 − ϕ0 )V0 . Now consider
deviation ϕ > ϕ0 . Then πid,r (ϕ, δi ) = β(1 − ϕ)V0 < β(1 − ϕ0 )V0 = πid,r (ϕ∗i , δi ). The inequality is
from ϕ > ϕ0 .
 Oh, Glaeser, and Su: Food Ordering and Delivery
38 Article submitted to Management Science

(b) If βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βfmin (δi ):

In this case, ϕ∗i should satisfy βf (ϕ∗i , δi ) = fmin (δi ) + ∆f ∗ , and such ϕ∗i gives πid,r (ϕ∗i , δi ) =
β(1 − ϕ∗i )V0 . Consider deviation ϕ < ϕ∗i . Then πid,r (ϕ, δi ) = (1 − ϕ)V0 ≤ (1 − ϕ0 )V0 = β(1 −
ϕmax (δi ))V0 < β(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ). The first equality holds because bidding less than ϕ∗i
will not give premium slot allocation. The first inequality holds because ϕ ≥ ϕ0 . The second equal-
ity is from Lemma 2. The second inequality is from ϕ∗i < ϕmax (δi ) since βfmax (δi ) > fmin (δi ) +
∆f ∗ . Now consider deviation ϕ > ϕ∗i . Then πid,r (ϕ, δi ) = β(1−ϕ)V0 < β(1−ϕ∗i )V0 = πid,r (ϕ∗i , δi ).
(c) If fmin (δi ) + ∆f ∗ > βfmax (δi ):
This is when i ≥ K + 1 so that the restaurant bids ϕ∗i = ϕmin (δi ) = ϕ0 , and πid,r (ϕ∗i , δi ) =
(1 − ϕ0 )V0 . Consider deviation ϕ > ϕ0 . Note that since fmax (δi ) < fmin (δi ) + ∆f ∗ , the restau-
rant cannot afford to bid high enough for a premium display slot. Therefore, for any deviation
ϕ ∈ (ϕ0 , ϕmax (δi )], the restaurant will not gain premium display. Thus, πid,r (ϕ, δi ) = (1 − ϕ)V0 <
(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ).
From Cases (a) through (c) we have shown that there is no profitable deviation from ϕ∗i for restaurant-i
with δi ≤ x0 .
(2) δi ∈ (x0 , xα ]: The equilibrium bid ϕ∗i is characterized as the following.

∗
ϕi = ψ(δi ),
 if βfmin (δi ) > fmin (δi ) + ∆f ∗ ,
βf (ϕi , δi ) = fmin (δi ) + ∆f , if βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βfmin (δi ),
∗ ∗

ϕ∗ = ψ(δ ),

otherwise.
i i

(a) If βfmin (δi ) > fmin (δi ) + ∆f ∗ :

In this case, ϕ∗i = ϕmin (δi ) = ψ(δi ). Also, i ∈ {1, 2, · · · , K}, and bidding ϕ∗i = ψ(δi ) will give
πid,r (ϕ∗i , δi ) = β(1 − ψ(δi ))V0 . Now consider deviation ϕ > ψ(δi ). Then πid,r (ϕ, δi ) = β(1 −
ϕ)V0 < β(1 − ψ(δi ))V0 = πid,r (ϕ∗i , δi ). The inequality is from ϕ > ψ(δi ). Consider deviation ϕ <
ψ(δi ) with βf (ϕ, δi ) ≥ fmin (δi ) + ∆f ∗ . Then, πid,r (ϕ, δi ) = βα(1 − ϕ)V0 ≤ βα(1 − ϕ0 )V0 = β(1 −
ϕα )V0 ≤ β(1 − ψ(δi ))V0 = πid,r (ϕ∗i , δi ). The first equality holds from the fact that ψ(δi ) is the min-
imum bid required to receive low delivery fee. The first inequality holds from ϕ ≥ ϕ0 . The second
equality holds from the definition of ϕα . The second inequality holds from the fact that ψ(δi ) ≤ ϕα
for δi ≤ xα .
(b) If βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βfmin (δi ):
In this case, ϕ∗i should satisfy βf (ϕ∗i , δi ) = fmin (δi ) + ∆f ∗ , and such ϕ∗i gives πid,r (ϕ∗i , δi ) =
β(1 − ϕ∗i )V0 . Also, note that in this case, fmax (δi ) > f (ϕ∗i , δi ) ≥ fmin (δi ) so that ϕmax (δi ) > ϕ∗i ≥
ψ(δi ). Consider deviation ϕ with ϕ ∈ [ψ(δi ), ϕ∗i ). Then πid,r (ϕ, δi ) = (1 − ϕ)V0 ≤ (1 − ψ(δi ))V0 =
β(1 − ϕmax (δi ))V0 < β(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ). The first equality holds because bidding ϕ ∈
[ψ(δi ), ϕ∗i ) will give low delivery fee and no premium slot allocation. The first inequality holds
because ϕ ≥ ψ(δi ). The second equality is from Lemma 2. The second inequality is from ϕ∗i <
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 39

ϕmax (δi ). Consider deviation ϕ with ϕ ∈ [ϕ0 , ψ(δi )). Then πid,r (ϕ, δi ) = α(1 − ϕ)V0 ≤ α(1 −
ϕ0 )V0 = (1 − ϕα )V0 ≤ (1 − ψ(δi ))V0 = β(1 − ϕmax (δi ))V0 < β(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ). The first
equality holds because ϕ < ψ(δi ) will give high delivery fee. The first inequality holds because
ϕ ≥ ϕ0 . The second equality holds from the definition of ϕα . The second inequality holds because
ψ(δi ) ≤ ϕα for δi ≤ xα . The third equality holds from Lemma 2, and the third inequality holds
because ϕ∗i < ϕmax (δi ). Now, consider deviation ϕ > ϕ∗i . Then πid,r (ϕ, δi ) = β(1 − ϕ)V0 < β(1 −
ϕ∗i )V0 = πid,r (ϕ∗i , δi ).
(c) If fmin (δi ) + ∆f ∗ ≥ βfmax (δi ):
This is when i ≥ K + 1 so that the restaurant bids ϕ∗i = ϕmin (δi ) = ψ(δi ), and πr (ϕ∗i , δi ) = (1 −
ψ(δi ))V0 . Consider deviation ϕ < ψ(δi ). Then πid,r (ϕ, δi ) = α(1 − ϕ)V0 ≤ α(1 − ϕ0 )V0 = (1 −
ϕα )V0 ≤ (1 − ψ(δi ))V0 = πid,r (ϕ∗i , δi ). The first equality holds because bidding less than ψ(δi )
gives high delivery price. The first inequality holds because ϕ ≥ ϕ0 . The second equality is from
the definition of ϕα , and the second inequality holds because ψ(δi ) ≤ ϕα for δi ≤ xα . Consider
deviation ϕ > ψ(δi ). Note that since βfmax (δi ) ≤ fmin (δi ) + ∆f ∗ , the restaurant cannot afford to
bid high enough for a premium display slot. Therefore, for any deviation ϕ ∈ (ψ(δi ), ϕmax (δi )],
the restaurant will not gain premium display. Thus, πid,r (ϕ, δi ) = (1 − ϕ)V0 < (1 − ψ(δi ))V0 =
πid,r (ϕ∗i , δi ).
From Cases (a) through (c) we have shown that there is no profitable deviation from ϕ∗i for restaurant-i
with δi ∈ (x0 , xα ].
(3) δi ∈ (xα , xαβ ]: The equilibrium bid ϕ∗i is characterized as the following.
 ∗

ϕi = ϕ0 , if βfmin (δi ) > fmin (δi ) + ∆f ∗ ,
∗ ∗ ∗
βf (ϕi , δi ) = fmin (δi ) + ∆f , if βf (m(ψ(δi ), α), δi ) > fmin (δi ) + ∆f ≥ βfmin (δi ),



ϕ∗i = ψ(δi ), if βf (ψ(δi ), δi ) > fmin (δi ) + ∆f ∗ ≥ βf (m(ψ(δi ), α), δi )
βf (ϕ∗ , δi ) = fmin (δi ) + ∆f ∗ , if βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βf (ψ(δi ), δi )



 ∗ i


ϕi = ϕ0 , otherwise.

(a) If βfmin (δi ) > fmin (δi ) + ∆f ∗ :

In this case, ϕ∗i = ϕmin (δi ) = ϕ0 . Also, i ∈ {1, 2, · · · , K}, and bidding ϕ∗i = ϕ0 will give
πid,r (ϕ∗i , δi ) = βα(1 − ϕ0 )V0 . Consider deviation ϕ ∈ (ϕ0 , ψ(δi )). Then, πid,r (ϕ, δi ) = βα(1 −
ϕ)V0 < βα(1 − ϕ0 )V0 = πid,r (ϕ∗i , δi ). The inequality is from ϕ > ϕ0 . Consider deviation ϕ ≥ ψ(δi ).
Then, πid,r (ϕ, δi ) = β(1 − ϕ)V0 ≤ β(1 − ψ(δi ))V0 < β(1 − ϕα )V0 = βα(1 − ϕ0 )V0 = πid,r (ϕ∗i , δi ).
The first equality holds since ϕ ≥ ψ(δi ) will give delivery discount. The first inequality holds from
ϕ ≥ ψ(δi ). The second inequality holds since ψ(δi ) > ϕα when δi > xα . The second equality holds
from the definition of ϕα .
(b) If βf (m(ψ(δi ), α), δi ) > fmin (δi ) + ∆f ∗ ≥ βfmin (δi ):
The equilibrium bid in this case satisfies βf (ϕ∗i , δi ) = fmin (δi ) + ∆f ∗ , and ϕ∗i < m(ψ(δi ), α) <
 Oh, Glaeser, and Su: Food Ordering and Delivery
40 Article submitted to Management Science

ψ(δi ). Thus, πr (ϕ∗i , δi ) = βα(1 − ϕ∗i )V0 . Consider deviation ϕ < ϕ∗i . Then, πid,r (ϕ, δi ) = α(1 −
ϕ)V0 ≤ α(1 − ϕ0 )V0 = β(1 − ϕmax (δi ))V0 ≤ β(1 − ψ(δi ))V0 = βα(1 − m(ψ(δi ), α))V0 < βα(1 −
ϕ∗i )V0 = πid,r (ϕ∗i , δi ). The first equality holds because bidding less than ϕ∗i will not give premium
display. The first inequality is from ϕ ≥ ϕ0 , and the second equality is from Lemma 2. The second
inequality holds because ψ(δi ) ≤ ϕmax (δi ), and ψ(δi ) guarantees both low price and premium
display. The second equality holds from the definition of m(ψ(δi ), α), and the last inequality holds
since ϕ∗i < m(ψ(δi ), α). Consider deviation ϕ with ϕ ∈ (ϕ∗i , ψ(δi )). Then, πid,r (ϕ, δi ) = βα(1 −
ϕ)V0 < βα(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ). Consider deviation ϕ with ϕ ≥ ψ(δi ). Then, πid,r (ϕ, δi ) =
β(1 − ϕ)V0 ≤ β(1 − ψ(δi ))V0 = βα(1 − m(ψ(δi ), α))V0 < βα(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ). The
first equality holds since with ϕ ≥ ψ(δi ), the restaurant qualifies for both low price and premium
display. The first inequality holds since ϕ ≥ ψ(δi ). The remaining follows from what we have
already shown previously with the deviation ϕ < ϕ∗i .
(c) If βf (ψ(δi ), δi ) > fmin (δi ) + ∆f ∗ ≥ βf (m(ψ(δi ), α), δi ):
In this case, ϕ∗i = ψ(δi ), and such ϕ∗i gives πid,r (ϕ∗i , δi ) = β(1 − ψ(δi ))V0 . Also, note that in this
case, fmax (δi ) > f (ϕ∗i , δi ) > fmin (δi ), and ϕmax (δi ) > ϕ∗i > ϕ0 . Consider deviation ϕ with ϕ < ϕ∗i
and βf (ϕ, δi ) ≥ fmin (δi ) + ∆f ∗ . Then, πid,r (ϕ, δi ) = βα(1 − ϕ)V0 ≤ βα(1 − m(ψ(δi ), α))V0 =
β(1 − ψ(δi ))V0 = πid,r (ϕ∗i , δi ). The first equality holds because with ϕ < ψ(δi ) and βf (ϕ, δi ) ≥
fmin (δi ) + ∆f ∗ , the restaurant receives high delivery price and a premium display slot. The first
inequality follows from the fact that βf (ϕ, δi ) ≥ fmin (δi ) + ∆f ∗ ≥ βf (m(ψ(δi ), α), δi ). The
second equality is from the definition of m(ψ(δi ), α). Consider deviation ϕ with ϕ < ϕ∗i and
βf (ϕ, δi ) < fmin (δi ) + ∆f ∗ . Then πid,r (ϕ, δi ) = α(1 − ϕ)V0 ≤ α(1 − ϕ0 )V0 = β(1 − ϕαβ )V0 ≤
β(1 − ψ(δi ))V0 = πid,r (ϕ∗i , δi ). The first equality holds since the restaurant receives high delivery
price and no premium slot when ϕ < ϕ∗i and βf (ϕ, δi ) < fmin (δi ) + ∆f ∗ . The first inequality holds
since ϕ ≥ ϕ0 . The second equality holds from the definition of ϕαβ , and the second inequality
holds from the fact that ϕαβ ≥ ψ(δi ) for δi ∈ (xα , xαβ ]. Consider deviation ϕ with ϕ > ϕ∗i . Then,
πid,r (ϕ, δi ) = β(1 − ϕ)V0 < β(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ).
(d) If βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βf (ψ(δi ), δi ):
The equilibrium bid in this case satisfies βf (ϕ∗i , δi ) = fmin (δi ) + ∆f ∗ , and ψ(δi ) ≤ ϕ∗i < ϕmax (δi ).
Thus, πid,r (ϕ∗i , δi ) = β(1 − ϕ∗i )V0 . Consider deviation ϕ with ϕ < ψ(δi ). Then, πid,r (ϕ, δi ) = α(1 −
ϕ)V0 ≤ α(1 − ϕ0 )V0 = β(1 − ϕmax (δi ))V0 < β(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ). The first equality holds
because bidding less than ϕ∗i will not give premium display, and bidding less than ψ(δi ) gives
high price. The first inequality is from ϕ ≥ ϕ0 , and the second equality is from Lemma 2. The
second inequality holds because ϕ∗i < ϕmax (δi ). Consider deviation ϕ with ϕ ∈ [ψ(δi ), ϕ∗i ). Then,
πid,r (ϕ, δi ) = (1 − ϕ)V0 ≤ (1 − ψ(δi ))V0 < (1 − ϕα )V0 = α(1 − ϕ0 )V0 = β(1 − ϕmax (δi ))V0 <
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 41

β(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ). The first equality holds sine bidding less than ϕ∗i does not give pre-
mium display. The first inequality is from ϕ > ψ(δi ), and the second inequality is from ψ(δi ) <
ϕα for δi > xα . The second equality is from the definition of ϕα , and the third equality holds
from Lemma 2. The third inequality holds since ϕ∗i < ϕmax (δi ). Consider deviation ϕ > ϕ∗i . Then,
πid,r (ϕ, δi ) = β(1 − ϕ)V0 < β(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ).
(e) If fmin (δi ) + ∆f ∗ ≥ βfmax (δi ):
In this case, the restaurant does not qualify for a premium display slot in equilibrium, and ϕ∗i =
ϕmin (δi ) = ϕ0 . Consider deviation ϕ with ϕ ∈ (ϕ0 , ψ(δi )). Then, πid,r (ϕ, δi ) = α(1 − ϕ)V0 <
α(1 − ϕ0 )V0 = πid,r (ϕ∗i , δi ). Consider deviation ϕ ≥ ψ(δi ). Then, πid,r (ϕ, δi ) = (1 − ϕ)V0 ≤ (1 −
ψ(δi ))V0 < (1 − ϕα )V0 = α(1 − ϕ0 )V0 = πid,r (ϕ∗i , δi ). The first equality and the first inequality hold
since ϕ ≥ ψ(δi ). The second inequality holds since ψ(δi ) > ϕα for δi > xα . The second equality
holds from the definition of ϕα .
From Cases (a) through (e)) we have shown that there is no profitable deviation from ϕ∗i for restaurant-i
with δi ∈ (xα , xαβ ].
(4) δi > xαβ : The equilibrium bid ϕ∗i can be characterized as the following.

∗
ϕi = ϕ0 ,
 if βfmin (δi ) > fmin (δi ) + ∆f ∗ ,
βf (ϕ∗i , δi ) = fmin (δi ) + ∆f ∗ , if βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βfmin (δi ),
ϕ∗ = ϕ ,

otherwise.
i 0

(a) If βfmin (δi ) > fmin (δi ) + ∆f ∗ :

In this case, ϕ∗i = ϕmin (δi ) = ϕ0 . Also, i ∈ {1, 2, · · · , K}, and bidding ϕ∗i = ϕ0 will give
πid,r (ϕ∗i , δi ) = βα(1 − ϕ0 )V0 . Note that a restaurant with δi > xαβ cannot qualify for low delivery
price with any bid below ϕmax (δi ). Now consider deviation ϕ > ϕ∗i . Then πid,r (ϕ, δi ) = βα(1 −
ϕ)V0 < βα(1 − ϕ0 )V0 = πid,r (ϕ∗i , δi ).
(b) If βfmax (δi ) > fmin (δi ) + ∆f ∗ ≥ βfmin (δi ):
In this case, ϕ∗i should satisfy βf (ϕ∗i , δi ) = fmin (δi ) + ∆f ∗ , and such ϕ∗i gives πid,r (ϕ∗i , δi ) =
βα(1 − ϕ∗i )V0 . Also, note that in this case, ϕmax (δi ) > ϕ∗i ≥ ϕmin (δi ). Consider deviation ϕ with
ϕ < ϕ∗i . Then, πid,r (ϕ, δi ) = α(1 − ϕ)V0 ≤ α(1 − ϕ0 )V0 = βα(1 − ϕmax (δi ))V0 < βα(1 − ϕ∗i )V0 =
πid,r (ϕ∗i , δi ). The first equality holds since bidding less than ϕ∗i does not give premium page dis-
play. The first inequality holds from ϕ ≥ ϕ0 , and the second equality holds from Lemma 2. The
last inequality holds since ϕmax (δi ) > ϕ∗i . Consider deviation ϕ with ϕ > ϕ∗i . Then πid,r (ϕ, δi ) =
βα(1 − ϕ)V0 < βα(1 − ϕ∗i )V0 = πid,r (ϕ∗i , δi ).
(c) If fmin (δi ) + ∆f ∗ ≥ βfmax (δi ):
This is when i ≥ K + 1, and the restaurant bids ϕ∗i = ϕmin (δi ) = ϕ0 , and πid,r (ϕ∗i , δi ) = α(1 −
ϕ0 )V0 . Consider deviation ϕ > ϕ0 . Note that since βfmax (δi ) ≤ fmin (δi ) + ∆f ∗ , the restaurant
cannot afford to bid high enough for a premium display slot. Therefore, for any deviation ϕ > ϕ0 ,
 Oh, Glaeser, and Su: Food Ordering and Delivery
42 Article submitted to Management Science

the restaurant will not gain premium display. Thus, πid,r (ϕ, δi ) = α(1 − ϕ)V0 < α(1 − ϕ0 )V0 =
πid,r (ϕ∗i , δi ).
From Cases (a) through (c) we have shown that there is no profitable deviation from ϕ∗i for
restaurant-i with δi > xαβ .
From (1) to (4), there is no incentive for any of the restaurants to deviate from ϕ∗i and ϕ∗i is the unique
subgame perfect equilibrium. ■

Proof of Proposition 3 (i): The centralized firm’s pricing decision is to choose the delivery price pFi B (δi )
that maximizes f (δi ), where

V0 + VL − cδi ,
 if pi = VL ,
f (δi ) = α(V0 + VH − cδi ), if pi = VH ,

0, if pi = ∞.

(1) The optimal price for the platform is pFi B (δi ) = VL when (V0 + VL − cδi ) ≥ α(V0 + VH − cδi ) and
L −αVH
(V0 + VL − cδi ) ≥ 0. The first condition holds when δi ≤ V0
c
+ V(1−α)c , and the second condition
V0 +VL VL −αVH VL
holds when δi ≤ c
. Since (1−α)c
< c
, it is optimal for the platform to charge VL when
V0 VL −αVH FB
δi ≤ c
+ (1−α)c
, that is, when δi ≤ x .
(2) The optimal price for the platform is VH when α(V0 + VH − cδi ) > (V0 + VL − cδi ) and α(V0 +
L −αVH
VH − cδi ) ≥ 0. The first condition holds when δi > i Vc0 + V(1−α)c , and the second condition holds
when δi ≤ V0
c
+ VcH . Therefore, VH is the optimal price when δi ∈ (xF B , y F B ].
(3) The optimal price for the platform is ∞ when 0 > (V0 + VL − cδi ) and 0 > α(V0 + VH − cδi ).
The first condition holds when δi > V0
c
+ VcL , and the second condition holds when δi > V0
c
+ VcH .
The second condition is the one that is binding. Therefore, ∞ becomes the optimal price when
δi > V0
c
+ VcH , that is, when δi > y F B .
■

Proof of Proposition 3 (ii): Let κ denote the set of restaurants that are featured. Given the first-best delivery
prices shown in Proposition 3 (i), the centralized firm’s profit π c (κ) is given as
N
X X
c
π (κ) = f F B (δi ) + (βf F B (δi ) − f F B (δi ))
i=1 j∈κ
N
X X
= f F B (δi ) + (β − 1)f F B (δi ).
i=1 j∈κ

Therefore, the profit maximizing decision for the centralized firm is to feature K restaurants that have the
highest f F B (δi ) values. Recall that f F B (δi ) is monotone decreasing in the delivery distance δi . Therefore,
the first-best display policy is to feature the K closest restaurants. ■
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 43

Proof of Proposition 4: We will follow the same steps as those to prove the decentralized equilibrium.
That is, we use backward induction to prove the second stage platform’s decision first (Proposition 4 (ii)),
and then prove the restaurants’ best response ϕci under the coordinated system (Proposition 4 (i)).

Proof of Proposition 4 (ii):

(1) Second stage platform decision for pc∗
i (ϕi , δi ):

Given a commission rate ϕi , we first show that the platform chooses the delivery price pc∗
i (ϕi , δi )

according to the First Best strategy shown in Proposition 3. Given ϕi , the platform’s delivery price
pc∗
i (ϕi , δi ) should maximize its profit


βϕi (V0 + VL − cδi ), if featured and pi = VL ,

ϕ (V + V − cδ ),
i 0 L i if not featured and pi = VL ,
πic,p (pi , ϕi , δi ) =

βϕ i α(V 0 + VH − cδi ), if featured and pi = VH ,

ϕi α(V0 + VH − cδi ), if not featured and pi = VH .


(a) Given any commission ϕi , platform’s best pricing strategy is to choose pc∗
i = VL if βϕi (V0 +

VL − cδi ) ≥ ϕi α(V0 + VH − cδi ) and βϕi (V0 + VL − cδi ) ≥ 0 if featured; and ϕi (V0 + VL −
cδi ) ≥ ϕi α(V0 + VH − cδi ) and ϕi (V0 + VL − cδi ) ≥ 0 if not featured. Since β cancels out
in the case of a featured restaurant, the decision is independent of whether the restaurant is
L −αVH
featured or not. The first condition holds when δi ≤ V0
c
+ V(1−α)c , and the second condition
V0 +VL VL −αVH VL
holds when δi ≤ c
. Since (1−α)c
< c
, it is optimal for the platform to charge VL when
L −αVH
δi ≤ V0
c
+ V(1−α)c , that is, when δi ≤ xF B .
(b) Given any ϕi , the optimal price for the platform is VH when ϕi α(V0 + VH − cδi ) > ϕi (V0 +
VL −αVH
VL − cδi ) and ϕi α(V0 + VH − cδi ) ≥ 0. The first condition holds when δi > i Vc0 + (1−α)c
,
and the second condition holds when δi ≤ V0
c
+ VcH . Therefore, VH is the optimal price when
δi ∈ (xF B , y F B ].
(c) Given any ϕi , the optimal price for the platform is ∞ when 0 > ϕi (V0 + VL − cδi ) and 0 >
ϕi α(V0 + VH − cδi ). The first condition holds when δi > V0
c
+ VcL , and the second condition
V0 VH
holds when δi > c
+ c
. The second condition is the one that is binding. Therefore, ∞
becomes the optimal price when δi > V0
c
+ VcH , that is, when δi > y F B .
From (a), (b), and (c), we can conclude that the platform’s pricing strategy under the coordinating
contract coincides with the optimal pricing strategy of the centralized firm shown in Proposition 3
(i). Note also that the platform’s delivery pricing strategy is independent of the commission rate
ϕi each restaurant chooses.
Here, we introduce some notations as the following. Recall that the below notations are analo-
gous to the notations that were used to prove Proposition 2. Let ϕcmin (δi ) represent the commission
 Oh, Glaeser, and Su: Food Ordering and Delivery
44 Article submitted to Management Science

rate restaurant-i would pay when there is no competition for premium display slots. This repre-
c
sents the commission rate the restaurant would pay only to receive delivery discount. Let fmin (δi )
represent the associated platform’s profit when the restaurant chooses ϕcmin (δi ) and is not featured.
That is, (
c ϕcmin (δi )(V0 + VL − cδi ), if δi ≤ xF B ,
fmin (δi ) =
ϕmin (δi )α(V0 + VH − cδi ), if δi > xF B .
c

Since the platform’s delivery price is independent of the restaurants’ commission rates, there is
no incentive for any restaurant to pay an additional commission rate besides the base rate ϕ0 for
delivery discount under the coordinating contract. Therefore, ϕcmin (δi ) = ϕ0 for all i.
Let ϕcmax (δi ) represent the maximum commission a restaurant is willing to pay when it can win
a featured display slot. That is, ϕcmax (δi ) should satisfy

β(1 − ϕcmax (δi ))(V0 + pc∗ c c∗ c

i (ϕmax (δi ), δi ) − cδi ) = (1 − ϕmin (δi ))(V0 + pi (ϕmin (δi ), δi ) − cδi ).

Note that in the above equation pc∗

i is independent of the restaurant’s commission rate, and we can

drop the restaurant’s commission rate ϕi from the arguments of the best response delivery price as
pc∗ c∗ c∗ c∗
i (ϕi , δi ) = pi (δi ) for notational simplicity. Then, pi (ϕmax (δi ), δi ) = pi (ϕmin (δi ), δi ), and

V0 + pc∗ c c∗ c c∗
i (ϕmax (δi ), δi ) − cδi = V0 + pi (ϕmin (δi ), δi ) − cδi = V0 + pi (δi ) − cδi .

This implies β(1 − ϕcmax (δi )) = (1 − ϕcmin (δi )), and

1 1
ϕcmax (δi ) = 1 − (1 − ϕcmin (δi )) = 1 − (1 − ϕ0 ).
β β

c
We also define βfmax (δi ) as the platform’s profit when the restaurant chooses to pay ϕcmax (δi ) and
becomes featured. That is,
(
c ϕcmax (δi )(V0 + VL − cδi ), if δi ≤ xF B ,
βfmax (δi ) =
ϕcmax (δi )α(V0 + VH − cδi ), if δi > xF B .

We also introduce a notation f c (ϕi , δi )

(
c (V0 + VL − cδi ), if δi ≤ xF B ,
f (δi ) =
α(V0 + VH − cδi ), if δi > xF B ,

as the profit restaurant-i brings to the system without considering the demand boost β from being
featured. The profit of the system can be inflated by β when featured and the profit is shared
between the platform and the restaurant according to the commission rate ϕi .
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 45

(2) Second stage platform decision for premium display:

Given ϕi , the platform’s best response strategy is to display K restaurants that bring highest prof-
its. In other words, given any ϕi , from i = 1, 2, · · · , N , the unique best response strategy for the
platform is to allocate the premium display slots to K restaurants to maximize its total profit
N
X X
π c,p (ϕ, κ) = c
fmin (δi ) + {βϕj f c (δj ) − fmin
c
(δj )},
i=1 j∈κ

where κ is the set of K featured restaurants. Then, the platform features restaurants with the highest
βϕj f c (δj )−fmin
c
(δj ) values in equilibrium. Note that this follows from the fact that restaurants that
are not featured in equilibrium have no incentive to deviate from ϕcmin (δi ) and only the restaurants
that are featured in equilibrium will pay an extra commission rate ϕi ≥ ϕcmin (δi ).
From (1), we have shown that the platform’s delivery price strategy under coordinating contract coincides
with the First-Best pricing strategy of the centralized firm. From (2) and the equilibrium restaurant com-
mission rate under the coordinating contract that is to be shown below, we can conclude that the platform
features the K restaurants that are closest to the customers, which coincides with the First-Best premium
display strategy of the centralized firm. ■

Proof of Proposition 4 (i): We now show the first stage restaurant decision on ϕci :
(a) Existence of the equilibrium:
We show the existence of the equilibrium by constructing the restaurants’ equilibrium commission
rates and showing that there is no incentive for the restaurants to deviate from the commission rates.
For simplicity, let us define ∆f c as ∆f c = βfmax
c c
(δK+1 ) − fmin (δK+1 ), which is the maximum
profit increase the K + 1-th marginal restaurant can give to the platform when featured. Given that
the platform follows the best response strategy shown in Proposition 4 (ii), the equilibrium bid ϕci
under the coordinating contract can be characterized as the following.

c c c
ϕi = ϕ0 ,
 if βfmin (δi ) ≥ fmin (δi ) + ∆f c ,
βϕci f c (δi ) = fmin
c
(δi ) + ∆f c , if βfmax
c c
(δi ) ≥ fmin (δi ) + ∆f c > βfmin
c
(δi ),
ϕc = ϕ ,

otherwise.
i 0

c c
(i) If βfmin (δi ) ≥ fmin (δi ) + ∆f c :
In this case, the restaurant is much closer to the customers than the marginal (K +1)-th restau-
c c c
rant with βfmax (δi ) > βfmin (δi ) ≥ fmin (δi ) + ∆f c , and the restaurant still gets featured while
paying ϕ0 . Thus, bidding ϕci = ϕ0 will give the restaurant πic,r (ϕci , δi ) = β(1 − ϕ0 )f c (δi ).
Now consider deviation ϕ > ϕ0 . Then πic,r (ϕ, δi ) = β(1 − ϕ)f c (δi ) < β(1 − ϕ0 )f c (δi ) =
πic,r (ϕci , δi ). The inequality is from ϕ > ϕ0 . Therefore, ϕ0 uniquely maximizes the restaurant’s
profit, and there is no profitable deviation.
 Oh, Glaeser, and Su: Food Ordering and Delivery
46 Article submitted to Management Science

c c
(ii) If βfmax (δi ) ≥ fmin (δi ) + ∆f c > βfmin
c
(δi ):
In this case, bidding ϕci with βϕci f c (δi ) = fmin
c
(δi ) + ∆f c guarantees restaurant-i being fea-
tured since such commission rate keeps the (K + 1)-th marginal restaurant and other far-
ther away restaurants out of the bidding game for featured display slots. Thus, πic,r (ϕci , δi ) =
β(1 − ϕci )f c (δi ). Consider deviation ϕ with ϕ0 ≤ ϕ < ϕci . Then πic,r (ϕ, δi ) = (1 − ϕ)f c (δi ) ≤
(1 − ϕ0 )f c (δi ) = β(1 − ϕcmax (δi ))f c (δi ) < β(1 − ϕci )f c (δi ) = πic,r (ϕci , δi ). The first equality
holds because bidding less than ϕci will not give premium slot allocation. The first inequality
holds because ϕ ≥ ϕ0 . The second equality is from the definition of ϕcmax (δi ). The second
inequality is from ϕci < ϕcmax (δi ). Now consider deviation ϕ > ϕci . Then πic,r (ϕ, δi ) = β(1 −
ϕ)f c (δi ) < β(1 − ϕci )f c (δi ) = πic,r (ϕci , δi ). Therefore, ϕci uniquely maximizes the restaurant’s
profit, and there is no profitable deviation.
c
(iii) If fmin (δi ) + ∆f c > βfmax
c
(δi ):
c c
This is when i ≥ K + 1. Note that since fmax (δi ) < fmin (δi ) + ∆f c , the firm cannot afford to
bid high enough to win a premium display slot. Consider deviation ϕ ∈ (ϕ0 , ϕcmax (δi )]. Then,
πic,r (ϕ, δi ) = (1 − ϕ)f c (δi ) < (1 − ϕ0 )f c (δi ) = πic,r (ϕci , δi ). Therefore, there is no profitable
deviation, and ϕci uniquely maximizes restaurant-i’s profit.
From Case (i) through (iii) we have shown that there is no profitable deviation from ϕci for
restaurant-i, and ϕci forms an equilibrium.
(b) Uniqueness of the equilibrium:
Given that the platform allocates the premium display slots to the K restaurants with the highest
βϕi f c (δi ) − fmin
c
(δi ) values (as shown in (2) of the proof of Proposition 4 (ii)), K restaurants with
largest ∆fic = βfmax
c c
(δi ) − fmin (δi ) value are featured. We can show this by contradiction. Note
c
that ∆fic decreases in δi since ∆fic = βfmax c
(δi ) − fmin (δi ) = βϕβ f c (δi ) − ϕ0 f c (δi ) = (βϕβ −
ϕ0 )f c (δi ) = (β − 1)f c (δi ), where f c (δi ) decreases in δi . Therefore, we can now assume that the
restaurants are ordered in the way that δ1 < δ2 < · · · < δN .Below, we show that if restaurant-i and
restaurant-j with i < j choose ϕi and ϕj under which restaurant-j is featured but restaurant-i is
not, then ϕi and ϕj cannot form an equilibrium.
If restaurant-j is featured, then it means that restaurant-j found a commission rate ϕj that satis-
fies ϕj ≤ ϕcmax (δj ) where βϕj f c (δj ) − fmin
c
(δj ) is among the K highest. Then, consider deviation
ϕ̂i = ϕj − ϵ with small ϵ > 0 for restaurant-i. With such deviation, restaurant-i can win a featured
display slot. This is because for small enough ϵ, β ϕ̂i f c (δi )−fmin
c
(δi ) = (β ϕ̂i −ϕ0 )f c (δi ) > (βϕj −
ϕ0 )f c (δj ) = βϕj f c (δj ) − fmin
c
(δj ). The inequality holds because f c (δ) decreases in δ and δi < δj .
Note also that deviating to ϕ̂i is profitable for restaurant-i since πic,r (ϕ̂i , δi ) = β(1 − ϕ̂i )f c (δi ) >
β(1 − ϕcmax (δi ))f c (δi ) = (1 − ϕcmin (δi ))f c (δi ) = (1 − ϕ0 )f c (δi ) ≥ (1 − ϕi )f c (δi ) = πic,r (ϕi , δi ).
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 47

The first equality holds because the restaurant is featured when it deviates to ϕ̂i ; the next inequal-
ity holds because ϕ̂i < ϕj ≤ ϕβ = ϕcmax (δi ); the next equality is from the relationship between
ϕcmax (δi ) and ϕcmin (δi ); the next equality is from ϕcmin (δi ) = ϕ0 ; the next inequality is because
ϕi ≥ ϕ0 . Therefore, ϕ̂i is a profitable deviation for restaurant-i, and ϕi cannot be an equilibrium
strategy for restaurant-i.
Therefore, if restaurants’ equilibrium best response strategy exists, it has to be unique in the
way that K restaurants closest to the customers are featured.
From (a) and (b), we have shown that the equilibrium exists, and in the unique equilibrium, K closest
restaurants to the customers are featured. ■

Combining Proposition 4 (i) and Proposition 4 (ii), we have shown that under the coordinating contract,
the equilibrium delivery pricing and featured display strategies coincide with those under the centralized
system, and the system is coordinated.
 Oh, Glaeser, and Su: Food Ordering and Delivery
48 Article submitted to Management Science

Appendix B. Formal Analyses of Linear Customer Demand Model

Proposition 5 (Second Stage Equilibrium for Linear Demand Model) Given the commision rates ϕi ,
(i) the platform’s optimal pricing policy pl∗ i (ϕi , δi ) is given as

0,
 if δi ≤ ϕi Vc0 −a
pl∗
i (ϕi , δi ) =
a−ϕi V0 +cδi
2
, if δi ∈ [ ϕi Vc0 −a , ϕi Vc0 +a ],
if δi ≥ ϕi Vc0 +a ,

a,

(ii) the platform’s optimal display policy is to feature the K restaurants that bring the platform the highest
profits under pl∗
i (ϕi , δi ).

pi


Proof of Proposition 5 (i): Optimal delivery price pl∗
i (ϕi , δi ) maximizes 1 − a (ϕi V0 + pi − cδi ) given

ϕi , while pi ∈ [0a]. From the first order condition, the profit function is maximized at pi with

dπil,p (pi , ϕi , δi ) 2 a − ϕi V0 + cδi

= − pi + = 0,
dpi a a
a−ϕi V0 +cδi
or when pi = 2
. Then, consider the following cases.
a−ϕi V0 +cδi
(i): When −ϕi V0 + cδi ≤ −a: In this case, 2
≤ 0 so that the profit maximizing delivery price
pl∗
i (ϕi , δi ) = 0.
a−ϕi V0 +cδi
(ii): When −a ≤ −ϕi V0 + cδi ≤ a, then the critical point of the profit function satisfies 2
∈
a−ϕi V0 +cδi
[0, a], and the profit maximizing delivery price satisfies pl∗
i (ϕi , δi ) = 2
.
(iii): When a ≤ −ϕi V0 + cδi then the critical point of the platform profit function is greater than a, i.e.,
a−ϕi V0 +cδi
2
≥ a, under which the demand becomes zero. Therefore, the optimal price in this case is
pl∗
i (ϕi , δi ) = a.

Proof of Proposition 5 (ii): The platform’s optimal display decision in equilibrium is to feature K restau-
rants that bring the largest profit increase to the platform when featured. The formal proof of this statement
is provided in the proof of Proposition 6. ■

L EMMA 5. When there is no chance of being featured, each restaurant-i chooses the commission rate
ϕlmin (δi ) characterized as the following.
(i) When (1 − ϕ0 )V0 − 2a ≥ 0:

ϕ0 ,
 if δi ≤ ϕ0 Vc0 −a ,
ϕlmin (δi ) = a+cδi
V0
, if δi ∈ ( ϕ0 Vc0 −a , V0 −3a
c
],
 V0 −a+cδi ,
 V0 −3a V0 +a
if δi ∈ ( c , c ].
2V0
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 49

(ii) When (1 − ϕ0 )V0 − 2a < 0:

(
ϕ0 , if δi ≤ (2ϕ0 −1)V 0 +a
,
ϕlmin (δi ) = V0 −a+cδi
c
(2ϕ0 −1)V0 +a V0 +a
2V0
, if δi ∈ ( c
, c ].

Proof of Lemma 5: Restaurant-i chooses ϕi that maximizes its profit 1 − pai (1 − ϕi )V0 given that ϕi ∈

[0, 1] and that the platform chooses its profit maximizing delivery price pl∗
i (ϕi , δi ) shown as in Proposition 5

(i). Thus, we need to consider the different cases where the best response delivery price takes different
functional forms.
a+cδi ϕi V0 −a
I. V0
≤ ϕi (i.e., δi ≤ c
): In this case, pl∗
i (ϕi , δi ) = 0 as shown in Proposition 5 (i). Thus, the

restaurant’s profit function becomes (1 − ϕi )V0 , and its profit maximizing problem is to find ϕi
ϕi V0 −a a+cδi
that maximizes (1 − ϕi )V0 while δi ≤ c
(or ϕi ≥ V0
) and ϕi ∈ [ϕ0 , 1]. In this case, there
are two possible cases where the feasible set is non-empty.
a+cδi ϕ0 V0 −a
(a) When V0
≤ ϕ0 (i.e., δi ≤ c
): Since the restaurant’s profit function is linear decreasing
in ϕi , the smallest ϕi in the feasible set becomes the optimal fee, and ϕlmin (δi ) = ϕ0 maximizes
the profit.
a+cδi ϕ0 V0 −a V0 −a

(b) When ϕ0 ≤ V0
≤ 1 (i.e., δi ∈ c
, c ): Then the smallest possible ϕi in the feasible
a+cδi
set that maximizes the profit function is ϕlmin (δi ) = V0
.
From (a) and (b) and the condition for I,
(
l ϕ0 , if δi ≤ ϕ0 Vc0 −a ,
ϕmin (δi ) = a+cδi ϕi V0 −a
V0
, if ϕ0 Vc0 −a ≤ δi ≤ c
.
cδi −a
II. V0
≤ ϕi ≤ cδi +a
V0
(i.e., δi ∈ ( ϕi Vc0 −a , ϕi Vc0 +a ]): In this case, the platform’s best strategy delivery
a−ϕi V0 +cδi
fee is pl∗
i (ϕi , δi ) = 2
as was shown in Propositions 5 (i), and the restaurant’s problem
becomes
 
a − ϕi V0 + cδi
maximizeϕi 1− (1 − ϕi )V0
2a
subject to ϕi ∈ [ϕ0 , 1],
 
−a + cδi a + cδi
ϕi ∈ , .
V0 V0

Note that the critical point, which maximizes the restaurant’s profit function 1 − a−ϕi2a
V0 +cδi


(1 −
V0 −a+cδi
ϕi )V0 is ϕi = 2V0
. Now, to understand the feasible region of the profit maximizing problem,
we should consider how the bounds of the two constraints compare to one another. There are four
−a+cδi
different possible orders among ϕ0 , 1, V0
and a+cδ
V0
i
, which determine the feasible set of the
−a+cδi a+cδi ϕ0 <−a+cδi a+cδi
restaurant’s profit maximization problem: (1) V0 < ϕ0 < V0
< 1, (2) V0
< V0
< 1,
 Oh, Glaeser, and Su: Food Ordering and Delivery
50 Article submitted to Management Science

−a+cδi a+cδi −a+cδi a+cδi

(3) V0
< ϕ0 < 1 < V0
, (4) ϕ0 < V0
<1< V0
.
We can rewrite the above conditions
as the following by rearranging the terms: (1) ϕ0 Vc0 −a < δi < min ϕ0 Vc0 +a , V0c−a , (2) ϕ0 Vc0 +a <


δi < V0c−a , (3) V0c−a < δi < ϕ0 Vc0 +a , (4) max V0c−a , ϕ0 Vc0 +a < δi < V0 +a

c
. There are two cases to
consider based on the parameter values:
ϕ0 V0 +a V0 −a
(A) c
≤ c
(i.e., (1 − ϕ0 )V0 − 2a ≥ 0): In this case, condition (3) is infeasible. Conditions
ϕ0 V0 −a
(1), (2), and (4) define three different regions for δi as the following: (1) c
< δi <
ϕ0 V0 +a
c
, (2) ϕ0 Vc0 +a < δi < V0c−a , and (4) V0c−a < δi < V0c+a .
(1) ϕ0 Vc0 −a < δi < ϕ0 Vc0 +a :
i. V0 −a+cδ
2V0
i
≥ a+cδ
V0
i
(i.e., δi ≤ V0 −3a
c
), then the critical point of the restaurant profit
function is greater than the upper bound of condition II, and the optimal fee for the
a+cδi
restaurant is at the upper bound of condition II, that is, ϕlmin (δi ) = V0
.
ii. ϕ0 ≤ V0 −a+cδi
2V0
≤ a+cδi
V0
(i.e., δi ≥ max{ (2ϕ0 −1)V
c
0 +a V0 −3a
, c }, then the critical point
of the restaurant profit function is between the lower bound and the upper bound of
the feasible region, and the optimal fee for the restaurant is the critical point of the
V0 −a+cδi
restaurant’s profit function, that is, ϕlmin (δi ) = 2V0
.
V0 −a+cδi (2ϕ0 −1)V0 +a
iii. 2V0
≤ ϕ0 (i.e., δi ≤ c
), then the critical point of the restaurant’s
profit function is less than ϕ0 , and ϕ0 maximizes the profit. That is, ϕlmin (δi ) = ϕ0 .
V0 −3a ϕ0 V0 −a
Note that i. is feasible only if c
≥ c
(i.e., (1 − ϕ0 )V0 − 2a ≥ 0), which
V0 −3a (2ϕ0 −1)V0 +a
is true under the Case (A) we are considering. In this case, c
− c
=
2(1−ϕ0 )V0 −4a
c
= 2c {(1 − ϕ0 )V0 − 2a} ≥ 0. In addition, note that the upper bound of δi
under iii. (2ϕ0 −1)V
c
0 +a
is smaller than the lower bound of δi in Region (1) ϕ0 Vc0 −a since
(2ϕ0 −1)V0 +a 0 −2a
c
− ϕ0 Vc0 −a = (1−ϕ0 )V
c
≥ 0 in Case (A). Therefore, iii. is not feasible under
Case (A), Region (1).
(2) ϕ0 V0 +a
c
< δi < V0c−a :
i. V0 −a+cδ
2V0
i
≥ a+cδ
V0
i
(i.e., δi ≤ V0 −3a
c
): This is when the critical point of restaurant’s
profit function is greater than the upper limit of ϕi as given by condition II, which
a+cδi
is V0
. Therefore, the optimal ϕi is the upper bound and ϕlmin (δi ) = a+cδ V0
i
.
−a+cδi
ii. V0
≤ V0 −a+cδ
2V0
i
≤ a+cδ
V0
i
(i.e., V0 −3a
c
≤ δi ≤ V0c+a ): In this case, the critical point
of the restaurant profit function is within the feasible region characterized by con-
V0 −a+cδi
dition II. Therefore, the optimal fee is ϕlmin (δi ) = 2V0
.
V0 −a+cδi cδi −a V0 +a
iii. 2V0
≤ V0
(i.e., δi ≥ c
): This case is incompatible with Region (2).
Note that there are two possible outcomes dependent on the parameter values.
V0 −3a
First, c
≤ ϕ0 Vc0 +a (i.e., (1 − ϕ0 )V0 − 4a ≤ 0): In this case, i. is infeasible and
ϕlmin (δi ) = V0 −a+cδ
2V0
i
for ϕ0 Vc0 +a < δi < V0c−a .
Second, V0 −3ac
≥ ϕ0 Vc0 +a (i.e., (1 − ϕ0 )V0 − 4a ≥ 0): In this case,
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 51
(
a+cδi
, if ϕ0 V0 +a
≤ δi ≤ V0 −3a ,
ϕlmin (δi ) = V0
V0 −a+cδi
c
V0 −3a
c
V0 −a
2V0
, if c
≤ δi ≤ c .
V0 −a V0 +a
(4) c
≤ δi ≤ c
:
cδi −a V0 −a+cδi
The only feasible case under this region is when V0
≤ 2V0
≤ 1, and ϕlmin (δi ) =
V0 −a+cδi
2V0
.
From Regions (1) through (4) under Case (A), we can conclude that
(
a+cδi
, if ϕ0 Vc0 −a ≤ δi ≤ V0 −3a ,
ϕlmin (δi ) = V0V−a+cδ
0
V0 −3a
c
V0 +a
2V0
i
, if c ≤ δi ≤ c .
ϕ0 V0 +a V −a
(B) c
> c
(i.e., (1 − ϕ0 )V0 − 2a < 0): In this case, condition (2) is infeasible, and condi-
ϕ0 V0 −a V0 −a
tions (1), (3), and (4) define three regions for δi as the following: (1) c
< δi < c
, (3)
V0 −a ϕ0 V0 +a
c
< δi < c
, (4) ϕ0 Vc0 +a < δi < V0c+a .
ϕ0 V0 −a
(1) c
< δi < V0c−a :
i. V0 −a+cδ
2V0
i
≥ a+cδ
V0
i
(i.e., δi ≤ V0 −3ac
): Note that condition i. cannot hold within region
(1) under Case (B) since V0 −3a c
− ϕ0 Vc0 −a = 1c {(1 − ϕ0 )V0 − 2a} < 0.
n o
V0 −3a (2ϕ0 −1)V0 +a
ii. ϕ0 ≤ V0 −a+cδ
2V0
i
≤ a+cδi
V0
(i.e., δ i ≥ max c
, c
): In this case, the crit-
ical point of the restaurant profit function is within the feasible region character-
V0 −a+cδi
ized by condition II. Therefore, the optimal fee is ϕlmin (δi ) = 2V0
. Also, note
(2ϕ0 −1)V0 +a (2ϕ0 −1)V0 +a
that condition ii. can be rewritten as δi ≥ c
since c
− V0 −3a
c
=
−1)V
− 2c {(1 − ϕ0 )V0 − 2a} > 0, and 0 c 0 − ϕ0 Vc0 −a = − 1c {(1 − ϕ0 )V0 − 2a ≥ 0.
(2ϕ +a

Condition ii. is feasible under Region (1) when (2ϕ0 −1)Vc

0 +a
< V0c−a , that is, when
V0 − a − {(2ϕ0 − 1)V0 + a} = 2{(1 − ϕ0 )V0 − a} ≥ 0.
V0 −a+cδi (2ϕ0 −1)V0 +a
iii. 2V0
≤ ϕ0 (i.e., δi ≤ c
): In this case, the critical point of the restau-
rant’s profit function is less than ϕ0 , the lower bound of the restaurant’s feasible
region defined for condition II, Case (B). Therefore, the optimal fee is ϕlmin (δi ) =
ϕ0 .
To summarize, the equilibrium restaurant fee for Region (1) is given as the following.
If (1 − ϕ0 )V0 − a ≥ 0, then
(
ϕ0 , if ϕ0 V0 −a
≤ δi ≤ (2ϕ0 −1)V 0 +a
,
ϕlmin (δi ) = V0 −a+cδi
c
(2ϕ0 −1)V0 +a
c
V0 −a
2V0
, if c
≤ δi ≤ c .

If (1 − ϕ0 )V0 − a ≤ 0, then
ϕlmin (δi ) = ϕ0
ϕ0 V0 −a
for c
< δi < V0c−a .
V0 −a
(3) c
≤ δi ≤ ϕ0 Vc0 +a :
 Oh, Glaeser, and Su: Food Ordering and Delivery
52 Article submitted to Management Science

V0 −a+cδi (2ϕ0 −1)V0 +a V0 +a

i. ϕ0 ≤ 2V0
≤ 1 (i.e., c
≤ δi ≤ c
): In this case, the critical point of
the restaurant profit function is within the feasible region characterized by condi-
V0 −a+cδi
tion II, Region (3). Therefore, the optimal fee is ϕlmin (δi ) = 2V0
.
V0 −a+cδi (2ϕ0 −1)V0 +a
ii. 2V0
≤ ϕ0 (i.e., δi ≤ c
): In this case, the critical point of the restau-
rant’s profit function is less than ϕ0 , the lower bound of the restaurant’s feasible
region defined for condition II, Case (B). Therefore, the optimal fee is ϕlmin (δi ) =
ϕ0 .
(2ϕ0 −1)V0 +a V0 −a
Depending on how the thresholds c
and c
compare to each other, we can
summarize the equilibrium restaurant fee as the following for Region (3).
(2ϕ0 −1)V0 +a V0 −a
If (1 − ϕ0 )V0 − a ≥ 0 (i.e., c
≤ c
), then

V0 − a + cδi
ϕlmin (δi ) = ,
2V0
V0 −a ϕ0 V0 +a
for c
≤ δi ≤ c
.
(2ϕ0 −1)V0 +a V0 −a
If (1 − ϕ0 )V0 − a ≤ 0 (i.e., c
≥ c
)
(
ϕ0 , if V0 −a
≤ δi ≤ (2ϕ0 −1)V 0 +a
,
ϕlmin (δi ) = V0 −a+cδi
c
(2ϕ0 −1)V0 +a
c
ϕ0 V0 +a
2V0
, if c
≤ δi ≤ c .

ϕ0 V0 +a V0 +a
(4) c
≤ δi ≤ c
:
V0 −a+cδi
Note that for this region, the critical point of the restaurant’s profit function 2V0
is
an interior point of the feasible region. We can see this as the lower bound of the feasible
−a+cδi V0 −a+cδi V +a V0 −a+cδi V +a
region V0
≤ 2V0
from δi ≤ c
. Also, 2V0
≤ 1 is true from δi ≤ c
.
V −a+cδ
Therefore, ϕlmin (δi ) = 0 2V0 i .
From Regions (1) through (4), we can conclude that under Case (B),
(
l ϕ0 , if ϕ0 Vc0 −a ≤ δi ≤ (2ϕ0 −1)V
c
0 +a
,
ϕmin (δi ) = V0 −a+cδi (2ϕ0 −1)V0 +a ϕ0 V0 +a
2V0
, if c
≤ δ i ≤ c
.

cδi −a ϕi V0 +a
III. ϕi ≤ V0
(i.e., δi ≥ c
): In this case, pl∗
i (ϕi , δi ) = a, and there is no demand.

By putting together all of the different cases studied above, we see that
(A) When (1 − ϕ0 )V0 − 2a ≥ 0:

ϕ0 ,
 if δi ≤ ϕ0 Vc0 −a ,
ϕlmin (δi ) = a+cδ
V0
i
, if δi ∈ ( ϕ0 Vc0 −a , V0 −3a
c
],
 V0 −a+cδi , if δ ∈ ( V0 −3a , V0 +a ].

2V0 i c c

(B) When (1 − ϕ0 )V0 − 2a < 0:

(
ϕ0 , if δi ≤ (2ϕ0 −1)V 0 +a
,
ϕlmin (δi ) = c
V0 −a+cδi
2V0
, if δi ∈ ( (2ϕ0 −1)V
c
0 +a V0 +a
, c ].
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 53

■
For the following analyses, let us assume that (1 − ϕ0 )V0 > aβ . This assumes that the food revenue V0 is
reasonably large enough compared to the maximum delivery price that can be charged by the platform. We
make this assumption to simplify the cases we would have to consider.
For simplicity of exposition let us define some notations as the following for the remainder
of the analyses. Let plmin,i = pl∗ l l l∗ l l
i (ϕmin (δi ), δi ) and pmax,i = pi (ϕmax (δi ), δi ). Also, let ϕmin,i =
l,r
ϕlmin (δi ), and ϕlmax,i = ϕlmax (δi ). In addition, πmin,i = πil,r (pl∗ l l l,p
i (ϕmin (δi ), δi ), ϕmin (δi ), δi ) and πmin,i =

πil,p (pl∗ l l
i (ϕmin (δi ), δi ), ϕmin (δi ), δi ).

L EMMA 6. When there is no featured display slot to bid for, the equilibrium delivery fee and the resulting
restaurant and platform profits are as the following.
(i) When (1 − ϕ0 )V0 − 2a ≥ 0:

0,
 if δi ≤ ϕ0 Vc0 −a ,
plmin,i = 0, if δi ∈ ( ϕ0 Vc0 −a , V0 −3a
c
],
 3a−V0 +cδi , if δ ∈ ( V0 −3a , V0 +a ].

4 i c c


(1 − ϕ0 )V0 ,
 if δi ≤ ϕ0 Vc0 −a ,
l,r
πmin,i = V0 − a − cδi , if δi ∈ ( ϕ0 Vc0 −a , V0 −3a
c
],
 (V0 +a−cδi )2
if δi ∈ ( V0 −3a , V0c+a ].

8a
, c

ϕ0 V0 − cδi ,
 if δi ≤ ϕ0 Vc0 −a ,
l,p
πmin,i = a, if δi ∈ ( ϕ0 Vc0 −a , V0 −3a
c
],
 (V0 +a−cδi )2
if δi ∈ ( V0 −3a , V0c+a ].

16a
, c

(ii) When (1 − ϕ0 )V0 − 2a < 0:


0,
 if δi ≤ ϕ0 Vc0 −a
plmin,i = a−ϕ0 V20 +cδi , if δi ∈ ( ϕ0 Vc0 −a , (2ϕ0 −1)V
c
0 +a
],
 3a−V0 +cδi , if δ ∈ ( (2ϕ0 −1)V0 +a , V0 +a ].

4 i c c


(1 − ϕ0 )V0 ,

 if δi ≤ ϕ0 Vc0 −a
l,r
πmin,i = a+ϕ0 V0 −cδi
2a
(1 − ϕ0 )V0 , if δi ∈ ( ϕ0 Vc0 −a , (2ϕ0 −1)V
c
0 +a
],
 (V0 +a−cδi )2 )
if δi ∈ ( (2ϕ0 −1)V 0 +a V0 +a

8a
, c
, c ].
ϕ0 V0 −a

ϕ0 V0 − cδi , 2
 if δi ≤ c
l,p (a+ϕ0 V0 −cδi ) ϕ0 V0 −a (2ϕ0 −1)V0 +a
πmin,i = 4a
, if δi ∈ ( c , c
],
 (V0 +a−cδi )2 )
 (2ϕ0 −1)V0 +a V0 +a
16a
, if δi ∈ ( c
, c ].
 Oh, Glaeser, and Su: Food Ordering and Delivery
54 Article submitted to Management Science

Proof of Lemma 6: The delivery fees follow from substituting ϕlmin,i we found from Lemma 5 to the
l,r
platform’s best response delivery fee pl∗
i (ϕi , δi ) function shown in Proposition 5 (i). The profits πmin,i and
l,p
πmin,i can be calculated by plugging ϕlmin,i and plmin,i into πil,p (pi , ϕi , δi ) = (1 − pai )+ (ϕi V0 + pi − cδi ) and
πil,r (pi , ϕi , δi ) = (1 − pai )+ (1 − ϕi )V0 . ■

L EMMA 7. The maximum commission rate restaurant-i is willing to pay for a featured display slot
ϕlmax (δi ) can be characterized as the following.
(i) When (1 − ϕ0 )V0 − 2a ≥ 0:


1 − β1 (1 − ϕ0 ), if δi ≤ ϕ0 Vc0 −a ,
if δi ∈ ( ϕ0 Vc0 −a , V0 −3a

1 − βV1 0 (V0 − a − cδi ), ],


l
c √
ϕmax (δi ) = 2 V0 +a−4aβ+4a β 2 −β
1 − (V0 +a−cδ i)
, if δi ∈ ( V0 −3a , ]
8aβV c √ c

 0

V0 +a−4aβ+4a β 2 −β V0 +a
q
 V0 −a+cδi + V0 +a−cδi 1 − 1 ,

if δi ∈ ( , c ].

2V0 2V0 β c

(ii) When (1 − ϕ0 )V0 − 2a < 0 and 4aβ{(1 − ϕ0 )V0 − a} > (1 − ϕ0 )2 V02 :


1
1 − β (1 − ϕ0 ),

 if δi ≤ ϕ0 Vc0 −a
a+ϕ V −cδ
(1 − ϕ0 ), if δi ∈ ( ϕ0 Vc0 −a , (2ϕ0 −1)V 0 +a

1 − β1 0 0 i
],


2a c √
ϕlmax (δi ) = 1 (V +a−cδ )2
(2ϕ −1)V +a V 0 +a−4aβ+4a β 2 −β
1− β 0 i
, if δi ∈ ( 0 0
, ],
8aV0 c √ c


 q 2 −β
 V −a+cδ V +a−cδ V +a−4aβ+4a β V +a
 0 i
+ 0 2V0 i 1 − β1 , if δi ∈ ( 0
, 0c ].

2V0 c

(iii) When (1 − ϕ0 )V0 − 2a < 0 and 4aβ{(1 − ϕ0 )V0 − a} < (1 − ϕ0 )2 V02 :



 1 − β1 (1 − ϕ0 ), if δi ≤ ϕ0 Vc0 −a
V0 −cδi
if δi ∈ ( ϕ0 Vc0 −a , δˆi ],

1 − β1 a+ϕ02a


(1 − ϕ0 ),


l
ϕmax (δi ) = V0 −a+cδi
q
4 (1−ϕ )V (ϕ V +a−cδ )
(V0 +a−cδi )2 − β 0 0 0 0 i


 2V0
+ 2V0
, if δi ∈ (δˆi , (2ϕ0 −1)V
c
0 +a
],
q
 V0 −a+cδi + V0 +a−cδi 1 − 1 , (2ϕ0 −1)V0 +a V0 +a

if δi ∈ ( , c ],

2V0 2V0 β c

where δˆi = 2aβ(V0 −a)−(a+ϕ0 V0 )V0 (1−ϕ0 )

{2aβ−(1−ϕ0 )V0 }c
.

Proof of Lemma 7: Recall that a restaurant’s maximum willingness pay ϕlmax (δi ) for the premium display
slot is the commission rate that leaves the restaurant indifferent between paying the lower bound commission
ϕlmin (δi ) while not being featured and paying the maximum commission while being featured. That is,
ϕlmax (δi ) satisfies

pl∗ l
 
i (ϕmax (δi ), δi )
β 1− (1 − ϕlmax (δi ))V0 = πil,r (pl∗ l l
i (ϕmin (δi ), δi ), ϕmin (δi ), δi ).
a

Proof of Lemma 7 (i): When (1 − ϕ0 )V0 − 2a ≥ 0:

Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 55

ϕ0 V0 −a l,r
a. δi ≤ c
: In this case, ϕlmin,i = ϕ0 , plmin,i = 0, and πmin,i = (1 − ϕ0 )V0 from Lemma 5 and
a+cδi
Lemma 6. Since ϕlmax,i ≥ ϕ0 and V0
≤ ϕ0 ≤ ϕlmax,i , the platform’s best response delivery fee
under ϕlmax,i is plmax,i = 0. Therefore, ϕlmax,i can be found by solving the following equation

β(1 − ϕlmax,i )V0 = (1 − ϕ0 )V0 ,

and ϕlmax,i = 1 − β1 (1 − ϕ0 ).
ϕ0 V0 −a V0 −3a a+cδi a+cδi
b. c
≤ δi ≤ c
: In this case, ϕlmin,i = V0
from Lemma 5. Since ϕlmax,i ≥ ϕlmin,i = V0
,
plmax,i = 0. Therefore, ϕlmax,i should satisfy

β(1 − ϕlmax,i )V0 = (V0 − a − cδi ),

l,r
where the right hand side of the equation follows from πmin,i = (V0 − a − cδi ). Rearranging the
equation gives ϕlmax,i = 1 − βV1 0 (V0 − a − cδi ).
V0 −3a V0 −a+cδi
c. δi ≥ c
: In this case, ϕlmin,i = 2V0
from Lemma 5. From Proposition 5 (i), we know that
there are two possible cases for plmax,i as the following.

if ϕlmax,i ≥ a+cδ
(
l
0, V0
i
,
pmax,i = a−ϕlmax,i V0 +cδi
2
, if −a+cδV0
i
≤ ϕlmax,i ≤ a+cδi
V0
.
a+cδi
case (1) When ϕlmax,i ≥ V0
: In this case, from Proposition 5 (i), plmax,i = 0, and ϕlmax,i should satisfy

(V0 + a − cδi )2
β(1 − ϕlmax,i )V0 = .
8a
l,r (V0 +a−cδi )2
Note that the right hand side of the equation comes from the fact that πmin,i = 8a
.
(1)
Let us label the ϕlmax,i under case (1) as ϕmax,i . Solving for the maximum commission rate
gives
(1) (V0 + a − cδi )2
ϕmax,i = 1 − .
8aβV0
n o
(1)
Note that ϕmax,i − a+cδ
d
dδi V0
i
= c(V04aβV
+a−cδi )
− Vc0 is a decreasing function in
n o 0 
(1)
δi and dδdi ϕmax,i − a+cδ V0
i
|δi =
V0 −3a =
1
β
− 1 c
V0
< 0. Also, when δi = V0 −3a c
,
n o c
(1) (1)
ϕmax,i − a+cδ
V0
i
|δ = V0 −3a = βV 2a
0
(β − 1) ≥ 0. Therefore, ϕmax,i − a+cδ
V0
i
> 0 at δi = V0 −3a
c
i c
V0 −3a V0 −3a
and decreases in δi for δi ≥ c
. Therefore, there exists a threshold δi′ ≥ c
such that for
(1) (1) (V0 +a−cδi )2
δi ∈ [ V0 −3a
c
, δi′ ], ϕmax,i is indeed greater than a+cδi
V0
and satisfies ϕmax,i = 1 − 8aβV0 with
plmax,i = 0.
−a+cδi a+cδi a−ϕlmax,i V0 +cδi
case (2) When V0
≤ ϕlmax,i ≤ V0
: In this case, from Proposition 5 (i), plmax,i = 2
,
and ϕlmax,i should satisfy

a − ϕlmax,i V0 + cδi (V0 + a − cδi )2

 
β 1− (1 − ϕlmax,i )V0 = .
2a 8a
 Oh, Glaeser, and Su: Food Ordering and Delivery
56 Article submitted to Management Science

(2)
Let us label such ϕlmax,i as ϕmax,i . Solving for the above quadratic equation gives
r
(2) V0 − a + cδi V0 + a − cδi 1
ϕmax,i = + 1− .
2V0 2V0 β

Now note that

r
(2) a + cδi V0 − 3a − cδi V0 + a − cδi 1
ϕmax,i − = + 1−
V0 2V0 2V0 β
q
is a decreasing function in δi and is equal to 1 − β1 > 0 at δi = V0 −3a
2a
V0 c
.
√
(1) (2) a+cδi V +a−4aβ+4a β(β−1)
Now note that ϕmax,i = ϕmax,i = V0
when δi = 0 c
. Therefore, we can confirm
that ϕlmax,i is given as
 √
ϕ(1) , if δ ∈ [ V0 −3a , V0 +a−4aβ+4a β(β−1) ],
i
ϕlmax,i = max,i c √ c
ϕ(2) , if δ ∈ [ V0 +a−4aβ+4a β(β−1) , V0 +a ]
max,i i c c

Summarizing a., b., and c. gives



 1 − β1 (1 − ϕ0 ), if δi ≤ ϕ0 Vc0 −a ,
if δi ∈ ( ϕ0 Vc0 −a , V0 −3a

1 − βV1 0 (V0 − a − cδi ), ],


l
c √
ϕmax (δi ) = 2 V0 +a−4aβ+4a β 2 −β
1 − (V0 +a−cδ i)
, if δi ∈ ( V0 −3a , ]
8aβV c √ c

 0

V0 +a−4aβ+4a β 2 −β V0 +a
q
 V0 −a+cδi + V0 +a−cδi 1 − 1 ,

if δi ∈ ( , c ].

2V0 2V0 β c

Proof of Lemma 7 (ii): When (1 − ϕ0 )V0 − 2a < 0 and 4aβ {(1 − ϕ0 )V0 − a} > (1 − ϕ0 )2 V02 :
ϕ0 V0 −a l,r
a. When δi ≤ c
: In this case, ϕlmin,i = ϕ0 , plmin,i = 0, and πmin,i = (1 − ϕ0 )V0 from Lemma 5
a+cδi
and Lemma 6. Since ϕlmax,i ≥ ϕ0 and V0
≤ ϕ0 ≤ ϕlmax,i , the platform’s best response delivery
fee under ϕlmax,i is plmax,i = 0. Therefore, ϕlmax,i can be found by solving the following equation

β(1 − ϕlmax,i )V0 = (1 − ϕ0 )V0 ,

and ϕlmax,i = 1 − β1 (1 − ϕ0 ).
ϕ0 V0 −a (2ϕ0 −1)V0 +a a+ϕ0 V0 −cδi l,r
b. When c
≤ δi ≤ c
: In this case, ϕlmin,i = ϕ0 , plmin,i = 2
, and πmin,i =
a+ϕ0 V0 −cδi


2a
(1 − ϕ0 )V0 from Lemma 5 and Lemma 6. From Proposition 5 (i), we know that there
are two possible cases for plmax,i as the following.

a+cδi
(
0, if ϕlmax,i ≥ V0
,
plmax,i = a−ϕlmax,i V0 +cδi −1+cδi a+cδi
2
, if V0
≤ ϕlmax,i ≤ V0
.
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 57

a+cδi
case (1) When ϕlmax,i ≥ V0
In this case, from Proposition 5 (i), plmax,i = 0, and ϕlmax,i should satisfy
:
 
l a + ϕ0 V0 − cδi
β(1 − ϕmax,i )V0 = (1 − ϕ0 )V0 .
2a
l,r
Note that the right hand side of the equation comes from the fact that πmin,i =
a+ϕ0 V0 −cδi

l (1)
2a
(1 − ϕ0 )V0 . Let us label the ϕmax,i under case (1) as ϕmax,i . Solving for the
maximum commission rate gives
 
(1) 1 a + ϕ0 V0 − cδi
ϕmax,i =1− (1 − ϕ0 ).
β 2a

for δi ∈ [ ϕ0 Vc0 −a , (2ϕ0 −1)V

(1) a+cδi 0 +a
Now, we need to show that ϕmax,i is indeed greater than V0 c
].
(1)
Note that ϕmax,i − a+cδ
V0
i
is a decreasing function in δi as
 
d (1) a + cδi c
ϕmax,i − = {(1 − ϕ0 )V0 − 2aβ} < 0.
dδi V0 2aβV0
(2ϕ0 −1)V0 +a
The inequality follows from (1 − ϕ0 )V0 − 2a < 0. Also, note that at δi = c
,

(1) a + cδi 2aβV0 − (1 − ϕ0 )2 V02 − 2aβ {2a + (2ϕ0 − 1)V0 }

ϕmax,i − =
V0 2aβV0
4aβ {(1 − ϕ0 )V0 − a} − (1 − ϕ0 )2 V02
= > 0,
2aβV0
where the inequality follows from the condition 4aβ {(1 − ϕ0 )V0 − a} > (1 − ϕ0 )2 V02 . There-
fore, for δi ∈ [ ϕ0 Vc0 −a , (2ϕ0 −1)V 0 +a (1)
c
], ϕmax,i in fact is the maximum commission rate.
a+cδi a−ϕlmax,i V0 +cδi
case (2) When ϕlmax,i < V0
: In this case, from Proposition 5 (i), plmax,i = 2
, and ϕlmax,i
should satisfy
a − ϕlmax,i V0 + cδi
   
a + ϕ0 V0 − cδi
β 1− (1 − ϕlmax,i )V0 = (1 − ϕ0 )V0 .
2a 2a
Solving for ϕlmax,i from the above equation gives
q
V0 − a + cδi (V0 + a − cδi )2 − β4 (1 − ϕ0 )V0 (ϕ0 V0 + a − cδi )
l
ϕmax,i = + .
2V0 2V0
(2) (2)
Let us label the ϕlmax,i under case (2) as ϕmax,i . Now, we need to check if ϕmax,i can be an
(2) a+cδi
eligible upper bound by checking if ϕmax,i − V0
< 0. If not, this case is not a sustainable
(2)
upper bound of the restaurant commission rate. First, note that ϕmax,i − a+cδ
V0
i
is a decreasing
function in δi in [ ϕ0 Vc0 −a , (2ϕ0 −1)V
c
0 +a
] since
 
d (2) a + cδi c cβ {−β(V0 + a − cδi ) + 2V0 (1 − ϕ0 )}
ϕmax,i − =− + ,
dδi V0 2V0 A
p
where A = β 2 (V0 + a − cδi )2 − 4βV0 (a + ϕ0 V0 − cδi )(1 − ϕ0 ) > 0, and −β(V0 + a −
(2ϕ0 −1)V0 +a
cδi ) + 2V0 (1 − ϕ0 ) is a linear increasing function in δi . Note that at δi = c
, −β(V0 +
 Oh, Glaeser, and Su: Food Ordering and Delivery
58 Article submitted to Management Science

a − cδi ) + 2V0 (1 − ϕ0 ) = 2(1 − β)(1 − ϕ0 )V0 , which is less than 0. Therefore, the difference
is a decreasing function in δi in [ ϕ0 Vc0 −a , (2ϕ0 −1)V (2ϕ0 −1)V0 +a
(2) 0 +a
ϕmax,i − a+cδ
V0
i
c
]. Also, at δi = c
,
p
(2) a + cδi 2a 4β 2 ϕ20 V02 − 4β(1 − ϕ0 )2 V02 − 4β 2 V02 (2ϕ0 − 1)
ϕmax,i − = ϕ0 − (2ϕ0 − 1) − + .
V0 V0 2βV0

From condition 4aβ {(1 − ϕ0 )V0 − a} > (1 − ϕ0 )2 V02 , we can see that the expression inside
the square root in the above can be shown to be

4β 2 ϕ20 V02 − 4β(1 − ϕ0 )2 V02 − 4β 2 V02 (2ϕ0 − 1)

≥ 4β 2 ϕ20 V02 − 16aβ 2 {(1 − ϕ0 )V0 − a} − 4β 2 V02 (2ϕ0 − 1)

= 4β 2 ϕ20 V02 − 16aβ 2 (1 − ϕ0 )V0 + 16a2 β 2 − 4β 2 V02 (2ϕ0 − 1)

= 4β 2 V02 (1 − ϕ0 )2 − 16aβ 2 (1 − ϕ0 )V0 + 16a2 β 2

= 4β 2 (1 − ϕ0 )2 V02 − 4a(1 − ϕ0 )V0 + 4a2



2
= 4β 2 {(1 − ϕ0 )V0 − 2a}

Therefore,
p
4β 2 ϕ20 V02 − 4β(1 − ϕ0 )2 V02 − 4β 2 V02 (2ϕ0 − 1)
q
2
≥ 4β 2 {(1 − ϕ0 )V0 − 2a}

= 2β{2a − (1 − ϕ0 )V0 },

(2ϕ0 −1)V0 +a
and at δi = c
,

(2) a + cδi 2a 2a − (1 − ϕ0 )V0

ϕmax,i − ≥ (1 − ϕ0 ) − +
V0 V0 V0
(1 − ϕ0 )V0 − 2a + 2a − (1 − ϕ0 )V0
= = 0.
V0

is a decreasing function in δi on [ ϕ0 Vc0 −a , (2ϕ0 −1)V

(2) 0 +a
Since ϕmax,i − a+cδ
V0
i
c
] and is non-negative
(2ϕ0 −1)V0 +a
is positive for all δi ∈ [ ϕ0 Vc0 −a , (2ϕ0 −1)V
(2) a+cδi 0 +a
at δi = c
, ϕmax,i − V0 c
]. This means
that case (2) is not an eligible upper bound of ϕli (δi ).
a+ϕ0 V0 −cδi


From case (1) and case (2), we can conclude that ϕlmax,i = 1 − β1 2a
(1 − ϕ0 ) and plmax,i =
0 when ϕ0 V0 −a
c
≤ δi ≤ (2ϕ0 −1)Vc
0 +a
.
(V0 +a−cδi )2
c. When δi ≥ (2ϕ0 −1)V
c
0 +a
: In this case, ϕlmin,i = V0 −a+cδi
2V0
, plmin,i = 3a−V0 +cδi
4
, l,r
and πmin,i = 8a

from Lemma 5 and Lemma 6. From Proposition 5 (i), we know that there are two possible cases
for plmax,i as the following.
a+cδi
(
0, if ϕlmax,i ≥ V0
,
plmax,i = a−ϕlmax,i V0 +cδi −a+cδi a+cδi
2
, if V0
≤ ϕlmax,i ≤ V0
.
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 59

a+cδi
case (1) When ϕlmax,i ≥ V0
: In this case, from Proposition 5 (i), plmax,i = 0, and ϕlmax,i should satisfy

(V0 + a − cδi )2
β(1 − ϕlmax,i )V0 = .
8a
l,r (V0 +a−cδi )2
Note that the right hand side of the equation comes from the fact that πmin,i = 8a
.
(1)
Let us label the ϕlmax,i under case (1) as ϕmax,i . Solving for the maximum commission rate
gives
(1) (V0 + a − cδi )2
ϕmax,i = 1 − .
8aβV0
n o
(1)
Note that dδdi ϕmax,i − a+cδ V0
i
= c(V04aβV
+a−cδi )
0
− Vc0 is a decreasing function in δi and
n o
(1) 0 )V0 −2aβ}
d
dδi
ϕmax,i − a+cδ
V0
i
|δ = (2ϕ0 −1)V0 +a = 2{(1−ϕ4aβV 0
< 0. Where the inequality comes
i c n o
from (1 − ϕ0 )V0 − 2a < 0. Also, at δi = (2ϕ0 −1)V 0 +a (1) a+cδi
c
, ϕ max,i − V0
|δ = (2ϕ0 −1)V0 +a =
i c
4aβ{(1−ϕ0 )V0 −a}−(1−ϕ0 )2 V0 (1) a+cδi
2aβV0
≥ 0. Therefore, ϕmax,i is indeed greater than V0
and satisfies
2
0 +a ˆ
with plmax,i = 0 for δi ∈ [ (2ϕ0 −1)V some threshold δˆi .
(1)
ϕmax,i = 1 − (V0 +a−cδ
8aβV0
i)
c
, δi ] for
−a+cδi a+cδi a−ϕlmax,i V0 +cδi
case (2) When V0
≤ ϕlmax,i ≤ V0
: In this case, from Proposition 5 (i), plmax,i = 2
,
and ϕlmax,i should satisfy

a − ϕlmax,i V0 + cδi (V0 + a − cδi )2

 
β 1− (1 − ϕlmax,i )V0 = .
2a 8a
(2)
Let us label such ϕlmax,i as ϕmax,i . Solving for the above quadratic equation gives
r
(2) V0 − a + cδi V0 + a − cδi 1
ϕmax,i = + 1− .
2V0 2V0 β
Now note that
r
(2) a + cδi V0 − 3a − cδi V0 + a − cδi 1
ϕmax,i − = + 1−
V0 2V0 2V0 β
(1) (2)
is a decreasing function in√δi . Also, from the proof of (i) c., we know
√that ϕmax,i = ϕmax,i =
a+cδi V0 +a−4aβ+4a β(β−1) (2ϕ0 −1)V0 +a V0 +a−4aβ+4a β(β−1) V0 +a
V0
at δi = c
where c
< c
< c
. There-
fore, we can confirm that ϕlmax,i
satisfies
 √
ϕ(1) , if δ ∈ [ (2ϕ0 −1)V0 +a , V0 +a−4aβ+4a β(β−1) ],
i
ϕlmax,i = max,i c √ c
ϕ(2) , if δ ∈ [ V0 +a−4aβ+4a β(β−1) , V0 +a ]
max,i i c c

Summarizing a., b., and c. gives



 1 − β1 (1 − ϕ0 ), if δi ≤ ϕ0 Vc0 −a
a+ϕ0 V0 −cδi
if δi ∈ ( ϕ0 Vc0 −a , (2ϕ0 −1)V0 +a


1 − β1 (1 − ϕ0 ), ],


2a c √
ϕlmax (δi ) = 2 V0 +a−4aβ+4a β(β−1)
1 − β1 (V0 +a−cδ i)
, if δi ∈ ( (2ϕ0 −1)V 0 +a
, ],
8aV0 c √ c


 q V0 +a−4aβ+4a β(β−1) V0 +a
 V0 −a+cδi + V0 +a−cδi 1 − 1 ,

if δi ∈ ( , c ].

2V0 2V0 β c
 Oh, Glaeser, and Su: Food Ordering and Delivery
60 Article submitted to Management Science

Proof of Lemma 7 (iii): When (1 − ϕ0 )V0 − 2a < 0 and 4aβ {(1 − ϕ0 )V0 − a} < (1 − ϕ0 )2 V02 :
ϕ0 V0 −a l,r
a. When δi ≤ c
: In this case, ϕlmin,i = ϕ0 , plmin,i = 0, and πmin,i = (1 − ϕ0 )V0 from Lemma 5
a+cδi
and Lemma 6. Since ϕlmax,i ≥ ϕ0 and V0
≤ ϕ0 ≤ ϕlmax,i , the platform’s best response delivery
fee under ϕlmax,i is plmax,i = 0. Therefore, ϕlmax,i can be found by solving the following equation

β(1 − ϕlmax,i )V0 = (1 − ϕ0 )V0 ,

and ϕlmax,i = 1 − β1 (1 − ϕ0 ).
b. When ϕ0 V0 −a
c
≤ δi ≤ (2ϕ0 −1)V
c
0 +a
: In this case, ϕlmin,i = ϕ0 , plmin,i = a+ϕ0 V20 −cδi , and πmin,i
l,r
=
a+ϕ0 V0 −cδi


2a
(1 − ϕ0 )V0 from Lemma 5 and Lemma 6. From Proposition 5 (i), we know that there
are two possible cases for plmax,i as the following.

if ϕlmax,i ≥ a+cδ
(
l
0, V0
i
,
pmax,i = a−ϕlmax,i V0 +cδi
2
, if −1+cδ
V0
i
≤ ϕlmax,i ≤ a+cδi
V0
.
a+cδi
case (1) When ϕlmax,i ≥ V0
In this case, from Proposition 5 (i), plmax,i = 0, and ϕlmax,i should satisfy
:
 
l a + ϕ0 V0 − cδi
β(1 − ϕmax,i )V0 = (1 − ϕ0 )V0 .
2a
l,r
Note that the right hand side of the equation comes from the fact that πmin,i =
a+ϕ0 V0 −cδi

(1)
2a
(1 − ϕ0 )V0 . Let us label the ϕlmax,i under case (1) as ϕmax,i . Solving for the
maximum commission rate gives
 
(1) 1 a + ϕ0 V0 − cδi
ϕmax,i =1− (1 − ϕ0 ).
β 2a
2aβ(V0 −a)−(a+ϕ0 V0 )V0 (1−ϕ0 ) (1)
Now, we need to show that for δi ≤ {2aβ−(1−ϕ0 )V0 }c
, ϕlmax,i = ϕmax,i . To do this,
(1) −a)−(a+ϕ0 V0 )V0 (1−ϕ0 )
we should show that ϕmax,i ≥ a+cδi
V0
for δi ∈ [ ϕ0 Vc0 −a , 2aβ(V0{2aβ−(1−ϕ 0 )V0 }c
]. Note that
(1)
ϕmax,i − a+cδ
V0
i
is a decreasing function in δi as
 
d (1) a + cδi c
ϕmax,i − = {(1 − ϕ0 )V0 − 2aβ} < 0.
dδi V0 2aβV0

The inequality follows from (1 − ϕ0 )V0 − 2a < 0. At δi = ϕ0 Vc0 −a ,

 
(1) a + cδi 1
ϕmax,i − = 1− (1 − ϕ0 ) > 0,
V0 β

Also, note that

(1) a + cδi {2aβV0 − (a + ϕ0 V0 − cδi )(1 − ϕ0 )V0 − 2aβ(a + cδi )}

ϕmax,i − =
V0 2aβV0
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 61

2aβV0 − (a + ϕ0 V0 )(1 − ϕ0 )V0 − 2a2 β − c {2aβ − (1 − ϕ0 )V0 } δi

= =0
2aβV0
2aβ(V0 −a)−(a+ϕ0 V0 )(1−ϕ0 )V0
when δi = c{2aβ−(1−ϕ0 )V0 }
. Note that

2aβ(V0 − a) − (a + ϕ0 V0 )(1 − ϕ0 )V0 (2ϕ0 − 1)V0 + a

−
c {2aβ − (1 − ϕ0 )V0 } c
2aβ(V0 − a) − (a + ϕ0 V0 )(1 − ϕ0 )V0 − {2aβ − (1 − ϕ0 )V0 } {(2ϕ0 − 1)V0 + a}
=
c {2aβ − (1 − ϕ0 )V0 }
4aβ {(1 − ϕ0 )V0 − a} − (1 − ϕ0 )2 V02
= < 0.
c {2aβ − (1 − ϕ0 )V0 }
Also,
2aβ(V0 − a) − (a + ϕ0 V0 )(1 − ϕ0 )V0 ϕ0 V0 − a 2a(β − 1)(1 − ϕ0 )V0
− = > 0.
c {2aβ − (1 − ϕ0 )V0 } c c {2aβ − (1 − ϕ0 )V0 }
a+cδi a−ϕlmax,i V0 +cδi
case (2) When ϕlmax,i < V0
: In this case, from Proposition 5 (i), plmax,i = 2
, and ϕlmax,i
should satisfy
a − ϕlmax,i V0 + cδi
   
l a + ϕ0 V0 − cδi
β 1− (1 − ϕmax,i )V0 = (1 − ϕ0 )V0 .
2a 2a
Solving for ϕlmax,i from the above equation gives
q
V0 − a + cδi (V0 + a − cδi )2 − β4 (1 − ϕ0 )V0 (ϕ0 V0 + a − cδi )
l
ϕmax,i = + .
2V0 2V0
n o
(2)
From the analysis of (ii) b. case (2), we know that dδdi ϕmax,i − a+cδ V0
i
is a decreasing func-
tion in δi when δi ∈ [ ϕ0 Vc0 −a , (2ϕ0 −1)V
c
0 +a
]. Note that

(1) a + cδi
ϕmax,i −
V0
q
V0 − 3a − cδi (V0 + a − cδi )2 − β4 (1 − ϕ0 )V0 (ϕ0 V0 + a − cδi )
= +
2V0 2V0
=0

when
4
(V0 − 3a − cδi )2 = (V0 + a − cδi )2 − (1 − ϕ0 )V0 (ϕ0 V0 + a − cδi ).
β
Rearranging the terms in the above equality gives
2aβ(V0 − a) − (a + ϕ0 V0 )(1 − ϕ0 )V0
δi = .
c {2aβ − (1 − ϕ0 )V0 }
From the analyses of case (1) and case (2), we can conclude that

1 − 1 a+ϕ0 V0 −cδi (1 − ϕ0 ), if δi ∈ [ ϕ0 Vc0 −a , δˆi ],


l β 2a
ϕmax,i = q
2 4
 V0 −a+cδi + (V0 +a−cδi ) − β (1−ϕ0 )V0 (ϕ0 V0 +a−cδi ) , if δi ∈ [δˆi , (2ϕ0 −1)V0 +a ],
2V0 2V0 c
 Oh, Glaeser, and Su: Food Ordering and Delivery
62 Article submitted to Management Science

and
if δi ∈ [ ϕ0 Vc0 −a , δˆi ],
(
0,
plmax,i = a−ϕlmax,i V0 +cδi
2
, if δi ∈ [δˆi , (2ϕ0 −1)V0 +a ].
c

where δˆi = 2aβ(V0 −a)−(a+ϕ0 V0 )(1−ϕ0 )V0

c{2aβ−(1−ϕ0 )V0 }
.
(2ϕ0 −1)V0 +a V0 −a+cδi 3a−V0 +cδi l,r (V0 +a−cδi )2
c. When δi ≥ c
: In this case, ϕlmin,i = 2V0
, plmin,i = 4
, and πmin,i = 8a

from Lemma 5 and Lemma 6. From Proposition 5 (i), we know that there are two possible cases
for plmax,i as the following.
a+cδi
(
0, if ϕlmax,i ≥ V0
,
plmax,i = a−ϕlmax,i V0 +cδi −a+cδi a+cδi
2
, if V0
≤ ϕlmax,i ≤ V0
.
a+cδi
case (1) When ϕlmax,i ≥ V0
: In this case, from Proposition 5 (i), plmax,i = 0, and ϕlmax,i should satisfy

(V0 + a − cδi )2
β(1 − ϕlmax,i )V0 = .
8a
l,r (V0 +a−cδi )2
Note that the right hand side of the equation comes from the fact that πmin,i = 8a
.
(1)
Let us label the ϕlmax,i under case (1) as ϕmax,i . Solving for the maximum commission rate
gives
(1) (V0 + a − cδi )2
ϕmax,i = 1 − .
8aβV0
n o
(1)
Note that dδdi ϕmax,i − a+cδ V0
i
= c(V04aβV
+a−cδi )
0
− Vc0 is a decreasing function in δi and
n o
(1) 0 )V0 −2aβ}
d
dδi
ϕmax,i − a+cδ
V0
i
|δ = (2ϕ0 −1)V0 +a = 2{(1−ϕ4aβV 0
< 0. Where the inequality comes
i c n o
(2ϕ0 −1)V0 +a (1) a+cδi
from (1 − ϕ0 )V0 − 2a < 0. Also, at δi = c
, ϕ max,i − V0
|δ = (2ϕ0 −1)V0 +a =
i c
4aβ{(1−ϕ0 )V0 −a}−(1−ϕ0 )2 V0 (1) a+cδi (1)
2aβV0
< 0. Therefore, ϕmax,i is less than V0
and ϕmax,i = 1 −
(V0 +a−cδi )2
8aβV0
cannot be the upper bound for ϕi .
−a+cδi a+cδi a−ϕlmax,i V0 +cδi
case (2) When V0
≤ ϕlmax,i ≤ V0
: In this case, from Proposition 5 (i), plmax,i = 2
,
and ϕlmax,i should satisfy
a − ϕlmax,i V0 + cδi (V0 + a − cδi )2
 
β 1− (1 − ϕlmax,i )V0 = .
2a 8a
(2)
Let us label such ϕlmax,i as ϕmax,i . Solving for the above quadratic equation gives
r
(2) V0 − a + cδi V0 + a − cδi 1
ϕmax,i = + 1− .
2V0 2V0 β
Now note that
r
(2) a + cδi V0 − 3a − cδi V0 + a − cδi 1
ϕmax,i − = + 1−
V0 2V0 2V0 β
(2ϕ0 −1)V0 +a
is a decreasing function in δi . Also, at δi = c
,
r
(2) a + cδi V0 − 3a − cδi V0 + a − cδi 1
ϕmax,i − = + 1−
V0 2V0 2V0 β
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 63
 
2(1 − ϕ0 )V0 1 − β1 − 4a
=
n 2V0  o
2 (1 − ϕ0 )V0 1 − β1 − 2a
= <0
2V0
Therefore, from case (1) and case (2), we know that
r
(2) V0 − a + cδi V0 + a − cδi 1 (2ϕ0 − 1)V0 + a V0 + a
ϕmax,i = + 1 − , if δi ∈ [ , ].
2V0 2V0 β c c
Summarizing a., b., and c. gives


 1 − β1 (1 − ϕ0 ), if δi ≤ ϕ0 Vc0 −a
V0 −cδi
if δi ∈ ( ϕ0 Vc0 −a , δˆi ],

1 − β1 a+ϕ02a


(1 − ϕ0 ),


l
ϕmax (δi ) = V0 −a+cδi
q
4 (1−ϕ )V (ϕ V +a−cδ )
(V0 +a−cδi )2 − β 0 0 0 0 i


 2V0
+ 2V0
, if δi ∈ (δˆi , (2ϕ0 −1)V
c
0 +a
],
q
 V0 −a+cδi + V0 +a−cδi 1 − 1 , if δi ∈ ( (2ϕ0 −1)V0 +a V0 +a

, c ],

2V0 2V0 β c

where δˆi = 2aβ(V0 −a)−(a+ϕ0 V0 )V0 (1−ϕ0 )

{2aβ−(1−ϕ0 )V0 }c
. ■

L EMMA 8. When the restaurants pay their maximum willingness to pay for featured display ϕlmax (δi ),
the equilibrium delivery fee is as the following.
(i) When (1 − ϕ0 )V0 − 2a ≥ 0:
 √
V0 +a−4aβ+4a β 2 −β
0, if δ i ≤ c √ ,
plmax,i = q
V +a−4aβ+4a β 2 −β
 3a−V0 +cδi − V0 +a−cδi 1 − 1 , if δi ∈ ( 0 , V0c+a ].
4 4 β c

(ii) When (1 − ϕ0 )V0 − 2a < 0 and 4aβ{(1 − ϕ0 )V0 − a} > (1 − ϕ0 )2 V02 :

 √
V0 +a−4aβ+4a β 2 −β
0, if δ i ≤ c √ ,
plmax,i = q 2
 3a−V0 +cδi − V0 +a−cδi 1 − 1 , if δi ∈ ( V0 +a−4aβ+4a β −β , V0 +a ].
4 4 β c c

(iii) When (1 − ϕ0 )V0 − 2a < 0 and 4aβ{(1 − ϕ0 )V0 − a} < (1 − ϕ0 )2 V02 :



 0, q
if δi ≤ δ̂2 ,
 2 4
(V0 +a−cδi ) − β (1−ϕ0 )V0 (ϕ0 V0 +a−cδi )
plmax,i = 3a−V40 +cδi −
q 4
, if δi ∈ (δ̂2 , (2ϕ0 −1)V
c
0 +a
],

 3a−V0 +cδi − V0 +a−cδi 1 − 1 , (2ϕ0 −1)V0 +a V0 +a
if δ ∈ (

4 4 β i ,c
], c

2aβ(V0 −a)−(a+ϕ0 V0 )V0 (1−ϕ0 )

where δ̂2 = {2aβ−(1−ϕ0 )V0 }c
.

Proof of Lemma 8: Substituting ϕlmax,i we found from Lemma 7 to the best response delivery price
pl∗ l
i (ϕi , δi ) found in Proposition 5 gives the pmax,i . ■
 Oh, Glaeser, and Su: Food Ordering and Delivery
64 Article submitted to Management Science

Proposition 6 (First Stage Equilibrium for Linear Demand Model) In equilibrium,

(i) the K restaurants with the highest virtual surplus ∆f l (δi ) choose the commission rates with ϕl∗
i ≥

ϕlmin (δi ) such that they match the virtual surplus of the (K + 1) − th restaurant. That is

ϕl∗ l
i = max{ϕmin (δi ), ϕ̃i },

where ϕ̃i is defined as

βπil,p (pl∗ l,p l∗ l l l

i (ϕ̃i , δi ), ϕ̃i , δi ) − πi (pi (ϕmin (δi ), δi ), ϕmin (δi ), δi ) = ∆f (δK+1 ).

(ii) all other restaurants choose commission rate ϕlmin (δi ).

Proof of Proposition 6: Using the same argument as Claims 1 through 3 in the proof of Proposition 2,
we can show that in a unique equilibrium, K restaurants with the highest virtual surplus are featured, and
no other structure of equilibrium can exist. We can also show that the equilibrium exists by showing that
there is no incentive for the restaurants and the platform to deviate from the above equilibrium as shown
in Proposition 6. The argument for the existence of the equilibrium is as the following. From the proofs of
Proposition 1 (ii) and Proposition 2 (i), we know that given ϕi from the restaurants, the platform should fea-
ture the K restaurants with the highest βπil,p (pl∗ l,p l∗ l l
i (ϕi , δi ), ϕi , δi ) − πi (p (ϕmin (δi ), δi ), ϕmin (δi ), δi ) values.

(i) Given the platform’s best response strategy, a restaurant with i ≤ K does not have any incentive to bid
ϕi > ϕ̃i . This is because increasing the bid does not change the platform’s decision, but it will only lower
the restaurant’s share of the food revenue. In addition, a restaurant with i ≤ K does not have any incentive
to bid ϕi ∈ [ϕlmin (δi ), ϕ̃i ). If the restaurant does bid less than ϕ̃i , then the (K + 1)th marginal restaurant can
bid ϕK+1 ∈ (ϕ̃i , ϕlmax (δK+1 )] and win restaurant-i for a featured display slot. This leaves restaurant-i worse
off because ϕ̃i ∈ [ϕlmin , ϕlmax ] so that for any ϕi ∈ [ϕlmin (δi ), ϕ̃i ),

πil,r (pl∗ (ϕi , δi ), ϕi , δi ) ≤ πil,r (pl∗ (ϕlmin , δi ), ϕlmin , δi ) = βπil,r (pl∗ (ϕlmax , δi ), ϕlmax , δi ).

The inequality comes from the optimality of ϕlmin , δi ) and the equality is by the definition of ϕlmin (δi ) and
ϕlmax (δi ), where ϕlmax (δi ) leaves the restaurant indifferent between being featured while paying ϕlmax (δi )
and not being featured while paying ϕlmin (δi ).
(ii) Given the platform’s best response strategy, a restaurant with i > K does not have any incentive to
unilaterally deviate from ϕlmin (δi ) as increasing the bid will help them win a featured display slot but will
only decrease its profit as ϕlmin (δi ) is the optimal fee when there is no featured display. ■

Proposition 7 (First Best for Linear Demand Model) For a centralized platform,
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 65

(i) the first-best pricing policy pl,F

i
B
is

0,
 if δi ≤ V0c−a ,
pl,F
i
B
= a−V0 +cδi
2
, if δi ∈ ( V0c−a , V0c+a ],
if δi > V0c+a .

a,

(ii) the first-best display policy is to feature the K closest restaurants.

Proof of Proposition 7 (i): The coordinated system’s problem of choosing the delivery price that maximizes
the total profit is to find pi ∈ [0, a] that maximizes
(
β 1 − pai (V0 + pi − cδi ), if featured,


l
πi (pi , δi ) =
1 − pai (V0 + pi − cδi ),


if not featured.

Note that the optimal delivery price policy maximizes β 1 − pai (V0 + pi − cδi ) if the restaurant is featured,

which is identical to maximizing 1 − pai (V0 + pi − cδi ). Therefore, as was the case with the two price

model, the First-best delivery price is independent of the platform’s featured display decision. Now, the
optimal delivery price should solve for the following profit maximization problem.
 pi 
max 1− (V0 + pi − cδi )
pi a
s.t. pi ∈ [0, a]
a−V0 +cδi
Note that the critical point of the quadratic profit function is pi = 2
.
V0 −a
(1) When δi ≤ c
: In this case, the critical point of the profit function is less than the lower bound
a−V0 +cδi
of the feasible region, that is, 2
≤ 0. Therefore, the optimal solution is the lower bound and
pl,F
i
B
= 0.
(2) When δi ∈ ( V0c−a , V0c+a ]: In this case, the critical point of the profit function is within the feasible
a−V0 +cδi
region of the optimization problem, that is, 2
∈ (0, a]. Therefore, the optimal solution is the
critical point, and pl,F
i
B
= a−V0 +cδi
2
.
V0 +a
(3) When δi > c
: In this case, the critical point of the profit function is greater than the upper bound
a−V0 +cδi
of the feasible region of the optimization problem, that is, 2
> a. Therefore, the optimal
solution is the upper bound, and pl,F
i
B
= a.
■

Proof of Proposition 7 (ii): Note that the optimal featured display policy is to feature the restaurants that
bring the highest profit to the system. By substituting the optimal delivery prices to the profit function of
the system gives the system profit f l,F B (δi ) as
 V0 −a
V0 − cδi , 2 if δi ≤ c ,

f l,F B (δi ) = (V0 +a−cδ
4a
i)
, if δi ∈ ( V0c−a , V0c+a ],
if δi > V0c+a .

0,

 Oh, Glaeser, and Su: Food Ordering and Delivery
66 Article submitted to Management Science

It is optimal to feature K restaurants with the highest f l,F B (δi ) values. Note that system profit f l,F B (δi ) is a
monotone decreasing and continuous function of δi . Therefore, under the First-best policy, the coordinated
system should feature K restaurants that are closes to the customers. ■

We can see that there is a difference between the equilibrium outcome derived for the decentralized
system shown in Proposition 5 and Proposition 6 and the first-best outcome shown in Proposition 7. First
of all, the delivery price for each restaurant in the decentralized system

0,
 if δi ≤ ϕi Vc0 −a
pl∗
i (ϕi , δi ) =
a−ϕi V0 +cδi
2
, if δi ∈ [ ϕi Vc0 −a , ϕi Vc0 +a ],
if δi ≥ ϕi Vc0 +a ,

a,

is larger than the first-best delivery price


0,
 if δi ≤ V0c−a ,
l,F B
pi = a−V0 +cδi
2
, if δi ∈ ( V0c−a , V0c+a ],
if δi > V0c+a .

a,

ϕi V0 −a
The free-delivery radius (service radius) in the decentralized system c
( ϕi Vc0 +a ) is smaller than that
V0 −a
in the first-best system c
( V0c+a ). This means that there are fewer restaurants that receive free delivery
(service) in the decentralized system compared to those in the centralized system. Also, within the region
a−ϕi V0 +cδi
where the delivery price is non-zero, the delivery price 2
is larger under the decentralized equilib-
a−V0 −cδi
rium than the price in the first-best system 2
. This is due to double marginalization as the platform
solely bears the delivery cost while sharing some fraction of the food revenue with the restaurant in the
decentralized system.
Another difference is found in the premium display policy. Recall that in the first-best outcome, the
K closest restaurants to the customers are being featured. However, under the decentralized equilibrium,
there are cases where farther-away restaurants overbid the closer-by restaurants and end up winning the
competition for premium display slots. We can see this from the fact that the virtual surplus ∆f l (δi ) is not
a monotone decreasing function of δi in the decentralized system. Since the platform ends up featuring the
K restaurants that have the largest virtual surplus ∆f l (δi ), non-decreasing ∆f l (δi ) means that the platform
features some far away restaurants that are inefficient to do so. In the following lemma, we show ∆f l (δi )
as a function of δi .

L EMMA 9. The virtual surplus ∆f l (δi ) under the decentralized system is given as the following.
(i) When (1 − ϕ0 )V0 − 2a ≥ 0:


(β − 1)(V0 − cδi ), if δi ≤ V0 −3a
c
, √
 2
V0 −3a V0 +a−4aβ+4a β 2 −β
∆f l (δi ) = β(V0 − cδi ) − 3(V0 +a−cδ i)
, if δ i ∈ ( , ]
16a c √ c

 (V0 +a−cδi )2
 √ 2 V0 +a−4aβ+4a β 2 −β V0 +a
8a
(β − 1 + β − β), if δi ∈ ( c
, c ].
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 67

(ii) When (1 − ϕ0 )V0 − 2a < 0 and 4aβ{(1 − ϕ0 )V0 − a} > (1 − ϕ0 )2 V02 :



(β − 1)(V0 − cδi ), if δi ≤ ϕ0 Vc0 −a
β(V − cδ ) − a+ϕ0 V0 −cδi (1 − ϕ )V − (a+ϕ0 V0 −cδi )2 , if δ ∈ ( ϕ0 V0 −a , (2ϕ0 −1)V0 +a ],


l 0 i 2a 0 0 4a i c c
∆f (δi ) = 3(V0 +a−cδi )2 (2ϕ0 −1)V0 +a
β(V 0 − cδi ) − 16a
, if δi ∈ ( c
, δ̂ 1 ],
 (V0 +a−cδi )2 (β − 1 + √β 2 − β),


 V0 +a
8a
if δ ∈ (δ̂ , ]. i 1 c

(iii) When (1 − ϕ0 )V0 − 2a < 0 and 4aβ{(1 − ϕ0 )V0 − a} < (1 − ϕ0 )2 V02 :

if δi ≤ ϕ0 Vc0 −a


(β − 1)(V0 − cδi ),
2
β(V0 − cδi ) − a+ϕ0 V0 −cδi (1 − ϕ0 )V0 − (a+ϕ0 V0 −cδi ) , if δi ∈ ( ϕ0 V0 −a , δ̂2 ],


2a 4a c
∆f l (δi ) = (V0 +a−cδi +A)2 (a+ϕ0 V0 −cδi )2 (2ϕ0 −1)V0 +a

β 8a
− q 4a
, if δi ∈ (δ̂ 2 , c
],
 V −a+cδ V +a−cδ (2ϕ −1)V +a
 0 i
+ 0 2V0 i 1 − β , 1
if δi ∈ ( 0 c 0 , 0c+a ], V

2V0

√
V0 +a−4aβ+4a β 2 −β
where the thresholds δ̂1 and δ̂2 satisfy δ̂1 = c
and
2aβ(V0 −a)−(a+ϕ0 V0 )V0 (1−ϕ0 )
δ̂2 = {2aβ−(1−ϕ0 )V0 }c
, and the term A is given as A =
q
(v0 + a − cδi )2 − β4 (1 − ϕ0 )V0 (ϕ0 V0 + a − cδi ).

Proof of Lemma 9: The virtual surplus ∆f l (δi ) can be found by

∆f l (δi ) = βfmax
l l
(δi ) − fmin (δi ),

while substituting ϕlmin,i , ϕlmax,i , plmin,i , and plmax,i found in Lemma 5, Lemma 6, Lemma 7, and Lemma 8
to +
pl∗ (ϕl (δi ), δi )

l
fmax (δi ) = 1 − i max (ϕlmax (δi )V0 + pl∗ l
i (ϕmax (δi ), δi ) − cδi ),
a
and +
pl∗ (ϕl (δi ), δi )

l
fmin (δi ) = 1 − i min (ϕlmin (δi )V0 + pl∗ l
i (ϕmin (δi ), δi ) − cδi ).
a

Proposition 8 With food revenue sharing and delivery cost splitting, the platform’s optimal pricing and
display policies coincide with the first-best policies.

Proof of Proposition 8: We will follow the same steps as the steps to prove the second best equilibrium.
That is, we use backward induction to prove the second stage platform’s decision first, and then prove the
restaurants’ best response ϕci under the coordinated system.
(1) Second stage platform decision: Given a commission rate ϕi , we first show that the platform
chooses the delivery price pl,c∗
i (ϕi , δi ) according to the First Best strategy shown in Proposition 7.

Given ϕi , the platform’s delivery price pl,c∗

i (ϕi , δi ) should maximize its profit
(
βϕi 1 − pai (V0 + pi − cδi ), if featured,


l,p,c
πi (pi , δi ) =
ϕi 1 − pai (V0 + pi − cδi ),


if not featured.
 Oh, Glaeser, and Su: Food Ordering and Delivery
68 Article submitted to Management Science

Note that the platform’s best response strategy is to choose pi that maximizes 1 − pai (V0 + pi −

cδi ) regardless of whether the restaurant is featured or not. Therefore, the profit maximization
problem becomes
 pi 
max 1− (V0 + pi − cδi )
pi a
s.t. pi ∈ [0, a]

which is identical to the profit maximization problem the centralized system solves to find the
first-best delivery price as shown in the proof of Proposition 7 (i). The resulting optimal delivery
price can be solved following the same steps of the proof of Proposition 7 (i), and

0,
 if δi ≤ V0c−a ,
pl,c
i =
a−V0 +cδi
2
, if δi ∈ ( V0c−a , V0c+a ],
if δi > V0c+a .

a,

The above delivery price is the same as the optimal delivery price for the centralized system pl,F
i
B

shown in Proposition 7 (i).

Note that the optimal delivery price chosen under the coordinating contract does not depend
on the restaurant commission ϕi . Therefore, the restaurants do not have an incentive to bid any
additional commission to free delivery and would pay the base commission rate ϕ0 if there were no
competition for featured display slots. Also, since the best response strategy for the delivery price
is independent of the commission rates restaurants are paying, the profit each restaurant brings to
the system (without being featured) can be written as
 V0 −a
V0 − cδi , 2 if δi ≤ c ,

f l,c (δi ) = (V0 +a−cδ
4a
i)
, if δi ∈ ( V0c−a , V0c+a ],
if δi > V0c+a ,

0,


l,c
pi (δi )
by substituting the pl,c
i obtained previously to the total welfare function (1 − a
)(V0 + pl,c
i (δi ) −

cδi ).
Now, we introduce some notations as the following. Let ϕl,c
min (δi ) represent the commission rate

restaurant-i would pay when there is no competition for premium display slots. From the fact that
the delivery fee is independent of the commission rate, we know that there is no incentive for the
restaurants to bid anything in addition to the base commission rate for the delivery discount. Thus,

ϕl,c
min (δi ) = ϕ0

l,p,c
for all of the restaurants. Let fmin (δi ) represent the associated platform’s profit when the restaurant
chooses ϕl,c l,c l,c
min (δi ). Thus, substituting ϕmin (δi ) = ϕ0 and pi gives

l,c,p
fmin (δi ) = ϕ0 f l,c (δi ).
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 69

Note that corresponding restaurant’s share is

l,c,r
fmin (δi ) = (1 − ϕ0 )f l,c (δi )

Let ϕl,c
max (δi ) represent the maximum commission a restaurant is willing to pay when it can win

a featured display slot. Since the optimal delivery fee under the coordinating contract does not
depend on the commission rate the restaurant chooses, ϕl,c
max (δi ) satisfies

l,c
β(1 − ϕl,c l,c l,c
max (δi ))f (δi ) = (1 − ϕmin (δi ))f (δi ),

that is,
l,c
β(1 − ϕl,c
max (δi )) = (1 − ϕmin (δi )),

1
which leads to ϕl,c
max (δi ) = ϕβ = 1 − β (1 − ϕ0 ). We introduce another notation here that represents

the platform and the restaurant’s profit associated the maximum commission rate (without being
featured) ϕl,c
max as
l,c,p
fmax (δi ) = ϕβ f l,c (δi ).

Note that corresponding restaurant’s share is

l,c,r
fmax (δi ) = (1 − ϕβ )f l,c (δi ).

(2) Following the same argument as in the proof of Proposition 4, given ϕi , the platform’s best
response strategy is to display K restaurants with the highest profit. In other words, given any ϕi ,
from i = 1, 2, · · · , N , the unique best response strategy for the platform is to allocate the premium
display slots to K restaurants to maximize its total profit
N
X X
l,c,p l,c,p
π l,c,p (ϕ, κ) = fmin (δi ) + {βϕj f l,c (δj ) − fmin (δj )},
i=1 j∈κ

where κ is the set of K featured restaurants. Then, the platform features restaurants with the
l,c,p
highest βϕj f l,c (δj ) − fmin (δj ) values in equilibrium. Note that this follows from the fact that
restaurants that are not featured in equilibrium have no incentive to deviate from ϕl,c
min (δi ) and only

the restaurants that are featured in equilibrium will pay an extra commission rate ϕi ≥ ϕc,l
min (δi ) =

ϕ0 .
(3) Now, we show that the restaurants that are featured under the coordinating contract coincides with
the restaurants that are featured under the first best outcome. That is, the K closest restaurants to
the customers are featured under the coordinating contract. We show this by following the same
steps of the proofs of Proposition 4.
(a) Claim 1: Given that the platform allocates the premium display slots to the K restaurants
l,c
with the highest βϕi f l,c (δi ) − fmin (δi ) values, in equilibrium, K restaurants with largest ∆fil,c =
 Oh, Glaeser, and Su: Food Ordering and Delivery
70 Article submitted to Management Science

l,c,p l,c,p
βfmax (δi ) − fmin (δi ) value are featured.
Proof: Note that ∆fil,c decreases in δi since ∆fil,c = βfmax
l,c,p l,c,p
(δi ) − fmin (δi ) = βϕβ f l,c (δi ) −
ϕ0 f l,c (δi ) = (βϕβ − ϕ0 )f l,c (δi ) = (β − 1)f l,c (δi ), where f l,c (δi ) decreases in δi . Therefore, we
can now assume that the restaurants are ordered in the way that δ1 < δ2 < · · · < δN . Below, we
show that if restaurant-i and restaurant-j with i < j choose ϕi and ϕj under which restaurant-j
is featured but restaurant-i is not, then ϕi and ϕj cannot form an equilibrium. If restaurant-j is
featured, then it means that restaurant-j found a commission rate ϕj that satisfies ϕj ≤ ϕl,c
max (δj )
l,c,p
where βϕj f l,c (δj ) − fmin (δj ) is among the K highest. Then, consider deviation ϕ̂i = ϕj − ϵ with
small ϵ > 0 for restaurant-i. With such deviation, restaurant-i can win a featured display slot. This
l,c,p
is because for small enough ϵ, β ϕ̂i f l,c (δi ) − fmin (δi ) = (β ϕ̂i − ϕ0 )f l,c (δi ) > (βϕj − ϕ0 )f l,c (δj ) =
l,c,p
βϕj f c (δj ) − fmin (δj ). The inequality holds because f l,c (δ) decreases in δ and δi < δj . Note also
that deviating to ϕ̂ is profitable for restaurant-i since πil,c,r (ϕ̂i , δi ) = β(1 − ϕ̂i )f l,c (δi ) > β(1 −
l,c l,c,r
ϕl,c l,c l,c l,c l,c
max (δi ))f (δi ) = (1 − ϕmin (δi ))f (δi ) = (1 − ϕ0 )f (δi ) ≥ (1 − ϕi )f (δi ) = πi (ϕi , δi ).
The first equality holds because the restaurant is featured when it deviates to ϕ̂; the next inequal-
ity holds because ϕ̂i < ϕj ≤ ϕβ = ϕl,c
max (δi ); the next equality is from the relationship between
l,c l,c
ϕl,c
max (δi ) and ϕmin (δi ); the next equality is from ϕmin (δi ) = ϕ0 ; the next inequality is because

ϕi ≥ ϕ0 . Therefore, ϕ̂i is a profitable deviation for restaurant-i, and ϕi cannot be an equilibrium

strategy for restaurant-i. Therefore, if restaurants’ equilibrium best response strategy exists, it has
to be unique in the way that K restaurants with shortest distance are featured.
(b) Claim 2: The equilibrium exists. We show the existence of the equilibrium by construct-
ing the restaurants’ equilibrium commission rates and showing that there is no incentive for the
l,c l,c
restaurants to deviate from the commission rates. For simplicity, let us define ∆fK+1 as ∆fK+1 =
l,c,p l,c,p
βfmax (δK+1 ) − fmin (δK+1 ), which is the maximum profit increase the K + 1-th marginal restau-
rant can give to the platform when featured. Then, the equilibrium bid ϕl,c
i under the coordinating

contract can be characterized as the following.


l,c l,c,p l,c,p
ϕi = ϕ0 ,
 if βfmin (δi ) ≥ fmin (δi ) + ∆f l,c (δK+1 ),
l,c l,c l,c,p l,c,p l,c,p
βϕi f (δi ) = fmin (δi ) + ∆f l,c (δK+1 ), if βfmax
l,c,p
(δi ) ≥ fmin (δi ) + ∆f l,c (δK+1 ) > βfmin (δi ),
ϕl,c = ϕ ,

otherwise .
i 0

l,c,p l,c,p
Case 1) βfmin (δi ) ≥ fmin (δi ) + ∆f l,c (δK+1 ): In this case, the restaurant is much closer to
l,c,p l,c,p l,c,p
the customers than the marginal (K + 1)-th restaurant with βfmax (δi ) > βfmin (δi ) ≥ fmin (δi ) +
∆f l,c (δK+1 ), and the restaurant still gets featured while paying ϕ0 . Thus, bidding ϕl,c
i = ϕ0

will give the restaurant πil,c,r (ϕl,c l,c

i , δi ) = β(1 − ϕ0 )f (δi ). Now consider deviation ϕ > ϕ0 . Then

πil,c,r (ϕ, δi ) = β(1 − ϕ)f l,c (δi ) < β(1 − ϕ0 )f l,c (δi ) = πil,c,r (ϕl,c
i , δi ). The inequality is from ϕ > ϕ0 .

Therefore, ϕ0 uniquely maximizes the restaurant’s profit, and there is no profitable deviation.
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 71

l,c,p l,c,p
l,c,p
Case 2) βfmax (δi ) ≥ fmin (δi ) + ∆f l,c (δK+1 ) > βfmin (δi ): In this case, bidding ϕl,c
i with
βϕl,c l,c l,c,p l,c
i f (δi ) = fmin (δi ) + ∆f (δK+1 ) guarantees restaurant-i being featured since such com-

mission rate keeps the (K + 1)-th marginal restaurant and other farther away restaurants out of
the bidding game for featured display slots. Thus, πil,c,r (ϕl,c l,c l,c
i , δi ) = β(1 − ϕi )f (δi ). Consider

deviation ϕ with ϕ0 ≤ ϕ < ϕl,c l,c,r

i . Then πi (ϕ, δi ) = (1 − ϕ)f l,c (δi ) ≤ (1 − ϕ0 )f l,c (δi ) = β(1 −
l,c l,c,r
ϕl,c l,c l,c
max (δi ))f (δi ) < β(1 − ϕi )f (δi ) = πi (ϕl,c
i , δi ). The first equality holds because bidding

less than ϕl,c

i will not give premium slot allocation. The first inequality holds because ϕ ≥ ϕ0 . The
l,c
second equality is from the definition of ϕl,c l,c
max (δi ). The second inequality is from ϕi < ϕmax (δi ).

Now consider deviation ϕ > ϕl,c l,c,r

i . Then πi (ϕ, δi ) = β(1 − ϕ)f l,c (δi ) < β(1 − ϕl,c l,c
i )f (δi ) =

πil,c,r (ϕl,c l,c

i , δi ). Therefore, ϕi uniquely maximizes the restaurant’s profit, and there is no profitable

deviation.
l,c,p
Case 3) fmin (δi ) + ∆f l,c (δK+1 ) > βfmax
l,c,p
(δ). This is when i ≥ K + 1. Note that since (δi ) <
l,c,p
fmin (δi ) + ∆f l,c (δK+1 ), the firm cannot afford to bid high enough to win a premium display slot.
l,c,r
Consider deviation ϕ ∈ (ϕ0 , ϕl,c
max (δi )]. Then, πi (ϕ, δi ) = (1 − ϕ)f l,c (δi ) < (1 − ϕ0 )f l,c (δi ) =
πil,c,r (ϕl,c l,c
i , δi ). Therefore, there is no profitable deviation, and ϕi = ϕ0 uniquely maximizes

restaurant-i’s profit.
From Case 1) through 3) we have shown that there is no profitable deviation from ϕl,c
i for
restaurant-i, and ϕl,c
i forms an equilibrium.

From the above arguments, we have shown that in the unique equilibrium, K closest restaurants
to the customers are featured.
Therefore, under the coordinating contract, the equilibrium delivery pricing and featured display
strategies coincide with those under the centralized system, and the system is coordinated. ■
 Oh, Glaeser, and Su: Food Ordering and Delivery
72 Article submitted to Management Science

Appendix C. Practical Implementation: Steps to Solve for the Equilibrium

In this appendix, we provide more details regarding the steps to find the equilibrium and the following
results for the model presented in Section 7.2. The steps to find the equilibrium mirror those of the base
model. Using backwards induction, the platform’s best response delivery price strategy is solved for as
shown in Section 7.2 and K restaurants that bring the highest profit boost from being featured win the
featured display slots. As we determine the K most profitable restaurants featured in the decentralized
equilibrium, we need to consider the virtual surplus, which depends on (1) the profit the restaurants can
bring to the platform when they pay their minimum commission and are not featured, and (2) the profit the
restaurants can bring to the platform when they are featured after paying the maximum commission rate
they are willing to pay for a featured display slot. (1) can be found by finding the commission rate ϕm (δ⃗i ) min

restaurant-i is willing to pay just for delivery discount and not for featured display. This would maximize
its aggregate profit without the competition for featured display where the aggregated profit Πr (ϕi , δ⃗i ) is i

given as
 
 X X 
Πri (ϕi , p⃗∗i (ϕi , δ⃗i ), δ⃗i ) = hi mj (1 − ϕi )V0 + mj ′ α(1 − ϕi )V0 .

j ′ :δij ′ ∈(x(ϕi ),y(ϕi )]

j:δij <x(ϕi )

Note that the superscript m represents “multiple market location,” δ⃗i = [δij ]j=1,2,·,M represents restaurant-
i’s distance vector to different market locations, and p⃗i is a vector of restaurant-i’s delivery price charged
to different market locations. Similar to what was done in Section 4.2, we can define the platform’s profit
Fmin (δ⃗i ) from restaurant-i associated with ϕm (δ⃗i ) as min

Fmin (δ⃗i ) = Πpi (ϕm ⃗ ⃗∗ m ⃗ ⃗ ⃗

min (δi ), pi (ϕmin (δi ), δi ), δi )

where
 
 X X 
Πpi (ϕi , p⃗∗i (ϕi , δ⃗i ), δ⃗i ) = hi mj (ϕi V0 + VL − cδij ) + mj ′ α(ϕi V0 + VH − cδij ′ ) .

j ′ :δij ′ ∈(x(ϕi ),y(ϕi )]

j:δij <x(ϕi )

We can also define restaurant-i’s maximum willingness to pay for a featured display slot ϕm ⃗
max (δi ) as the

ϕi that satisfies
βΠri (ϕi , p⃗∗i (ϕi , δ⃗i ), δ⃗i ) = Πri (ϕm ⃗ ⃗∗ m ⃗ ⃗ ⃗
min (δi ), pi (ϕmin (δi ), δi ), δi ),

and the platform profit Fmax (δ⃗i ) associated with ϕm ⃗

max (δi ) as

Fmax (δ⃗i ) = Πpi (ϕm ⃗ ⃗∗ m ⃗ ⃗ ⃗

max (δi ), pi (ϕmax (δi ), δi ), δi ).

The platform then ranks the restaurants by the virtual surplus ∆Fim = βFmax (δ⃗i ) − Fmin (δ⃗i ) and features
the K restaurants that have the highest ∆F m values. Restaurant-i that is featured bids ϕm∗ (δ⃗i ) ≥ ϕm (δ⃗i )
i i min

that maximizes Πri (ϕi , p⃗∗i (ϕi , δ⃗i ), δ⃗i ) while satisfying

βΠpi (ϕm∗ ⃗ ⃗∗ m∗ ⃗ ⃗ ⃗ p m ⃗ ⃗∗ m ⃗ ⃗ ⃗ m
i (δi ), pi (ϕi (δi ), δi ), δi ) − Πi (ϕmin (δi ), pi (ϕmin (δi ), δi ), δi ) ≥ ∆FK+1 .
Oh, Glaeser, and Su: Food Ordering and Delivery
Article submitted to Management Science 73

to keep the (K + 1)-th restaurant out of the competition for featured display slots; restaurant-i that is not
featured pays ϕm∗ (δ⃗i ) = ϕm (δ⃗i ). Note that in this model with heterogeneous restaurant demand, virtual
i min

surplus and the outcome of the bidding game for premium display depends on the restaurant demand index
hi . As restaurants with shorter distance to customer location δi tend to have advantage in winning a premium
display slot in the base model with single customer location, restaurants with high demand index are more
likely to have higher virtual surplus in equilibrium. In addition, as the analysis of this model suggests,
our modeling framework is flexible enough to admit other model variations that incorporate other types
of heterogeneity for any restaurant-customer pair (i, j). For example, the per-unit delivery cost c can be
heterogeneous for any restaurant-market pair and be replaced as cij ; the price sensitivity for delivery service
α can be heterogeneous and be replaced as αij .