NPD Notes:

Ideation

Develop

Test

Launch/Marketing Mix

Track

Superiority: From customer’s or users pov. Customers pay for the benefits
they get.

Me-Too or Tired New Products: one that fails to excite the customers.

Serendipity: discovering by accident.

• Select strategic arenas that are both attractive (large, growing

markets; weak competition; good opportunities to develop new
products) and where you can leverage your core competencies in
technology, marketing, and production to advantage.

Design Thinking

Lead Users

Ethnography

Innovation Generation Process:

WHO

Who participates in generating the innovation?

Innovation is not a solo activity — it’s a strategic choice about contributors.

From the diagram:

Types of “WHO” sources

 Lead Users – Early adopters with advanced needs; extremely valuable

for breakthrough innovations.

 Crowdsourcing – Many contributors; great for idea breadth and

problem solving.

 Open Innovation – Innovation across organizational boundaries

(partners, universities, suppliers).
  Top-Down – Leadership-driven initiatives (strong direction, strategic
alignment).

 Grassroots – Employee-driven innovation (bottom-up creativity,

autonomy).

Instructor Insight:
Choosing WHO determines idea quality, speed, and alignment.
Wrong “who” = wrong ideas.
Right “who” = fast, relevant innovation.

What type of innovation are you creating?

This dimension directs the target of innovation, which can radically

change methods, risks, and timelines.

Types of “WHAT”:

 Product – A new or improved physical/digital offering.

 Service – New service delivery model, new customer experience, etc.

 Process – Internal workflow & operational innovation.

 Business Model – Revenue model, value delivery, ecosystem changes

(highest impact).

 Type: Risk & Horizon – Classifying innovation based on risk and

timeframe:

o Horizon 1: incremental; existing market

o Horizon 2: adjacent; medium risk

o Horizon 3: breakthrough; high risk/high reward

Instructor Insight:
Your opportunity size and investor interest change dramatically depending
on WHICH “WHAT” you pick.
A product change is small.
A business model change is enormous.

HOW

How will the innovation be developed?

This reflects your methodology, structure, speed, and flexibility.

Types of “HOW”:
  Stage-Gate – Structured, sequential phases with approvals (good for
regulated industries).

 Spiral – Iterative cycles of refinement (often used in software).

 Lean – Build–Measure–Learn loops; minimal waste.

 Design Thinking – Human-centered approach; empathy → ideation →

prototyping → testing.

 Systematic Inventive Thinking (SIT) – Formal creativity techniques

using constraints.

Instructor Insight:
Choosing the wrong HOW causes:

 Slow development

 Overbuilding

 Misalignment with market

 Failed launches
Choosing the right HOW accelerates learning, reduces cost, and
produces a product people actually want.

a business model describes the rationale of how an organization creates,

delivers and captures value.” In their work, a business model is based on
nine building blocks, including: (1) the customer segments a company
serves; (2) a company’s value proposition(s); (3) the channels a company
uses to deliver its value proposition(s); (4) the customer relationships a
company establishes; (5) a company’s revenue streams; (6) the key
resources a company uses; (7) the key activities a company performs; (8) the
key 42 Handbook of research on new product development partnerships the
company establishes; and (9) the cost structure adopted.

Most well known are: (1) incremental versus radical innovations; (2)
sustaining versus disruptive innovations; and (3) core versus adjacent versus
transforma tional innovations.

Nagji and Tuff (2012, pp. 68–9) distinguish between three innovation
initiatives: core, adjacent and transformational innovations. Core innovations
“are efforts to make incremental changes to existing products and
incremental inroads into new markets.” Adjacent innovations “leverage
something the company does well into a new space.” Transformational
innova tions “are new offers – if not whole new businesses – to serve new
markets and customer needs.” The authors find that firms that outperform
their competitors dedicate 70 percent of their investments to core
innovations, 20 percent to riskier adjacent innovations, and only 10 percent
to transformational innovation “gambles.” Interestingly, the authors find that
these transformational innovations are most profitable.

A first demographic is firms’ own employee base in its broadest sense,

leading to grassroots innovation efforts. A second demographic are select
customers, so-called lead users, of a firm that experience a need for an
innovation before the mainstream market. By extension, a company can
crowdsource its innovations from a very large demographic of customers or
the general public at large, which may include technical experts from other
fields. Last, we review open innovation logic, which advocates direct
cooperation of firms with external experts or other firms rather than sending
out a call for ideas to a large audience, like grassroots or crowdsourcing
approaches advocate

The stage-gate process uses multifunctional teams that work on several

activities at the same time (Cooper 2008). This approach makes it
considerably more efficient than phased project planning, which reportedly
doubles the development time of projects (Cooper 1994). A firm increases
resource investment as the project passes through successive stages and
uncertainty decreases. The stage-gate methodology incorporates a go-or-kill
decision after each stage. This decision is based on the progress of the
project as evaluated against a predetermined set of criteria.

Several scholars have inventoried the main disadvantages of the stage-gate

method. O’Connor (1994) concludes that implementing the stage-gate
process is neither easy nor quick. Cooper (1994) concludes that the stage-
gate process: (1) may delay the progress of a project because each stage
has to be completed before the next can start; (2) may be too heavy for
small, low-risk projects; and (3) does not prioritize projects and, thus, does
not focus resources on projects with the highest priority.

The spiral development process starts at the center of the spiral, which
represents the inception of the project idea. The spiral process is divided into
four main parts (Boehm 1988). (1) Finding out the goal of the project and
identifying technical, legal and feasibil ity constraints. (2) Risk identification,
which is crucial to the process. Ideally, the biggest risk is resolved first,
followed by the next biggest risk and so forth. If a risk cannot be resolved,
the project may be cancelled. (3) Verifying the concept by requesting custom
ers’ feedback on a prototype. (4) The customer feedback determines
whether a project is cancelled, put through to another iteration of the spiral
process or is ready for market launch.

The lean innovation process is based on speed of execution and on deploying

minimum

viable products that can receive continuous customer feedback.

To do so, a design process goes through the three stages of inspiration,

ideation and implementation.

While customers can be a great resource for innovation, they may not always
know what they need or lack the imagination to help develop innovations
(Goldenberg et al. 2003). As opposed to many other processes, systematic
inventive thinking is characterized by “inside the box” thinking and listening
to the voice of the product rather than solely to that of the consumer.

Subtraction etc comes under this.

Grassroots Innovation
📍 Organic innovation coming from employees, teams, or internal
practitioners.

The arrows pointing upward in the triangle represent ideas rising from
people closer to the work or the customer.

Characteristics:

 High authenticity and relevance

 Employees identify real operational or customer pain points

 Strong for continuous improvement and process innovation

 Aligns with intrinsic motivation

 Risk: ideas may not match leadership strategy unless guided

Examples:

 Toyota’s Kaizen system → small, continuous improvements

 Google’s 20% time → Gmail, Google News

 Hackathons inside companies

Horizon 1

Known need × known solution → optimize and improve.

Horizon 2

New need OR new solution → expand and evolve.

Horizon 3

New need × new solution → invent and explore.

Altshuller and Goldenberg:

• In practice, the process starts by

1. listing the physical components and attributes of existing products,

 2. understanding the products’ direct environment, and

3. seeking paths for improvement

• Innovation can then follow one or more of five identified patterns.

• Subtraction entails removing undesirable components from the product

in question.

• Multiplication involves creating copies of certain product components

and then changing those copied parts to develop an innovation.

• This is what Gillette did to its razor when it added a second blade at a
different angle to the first, allowing the first blade to raise the whiskers
and the second to achieve a clean shave.

• Division entails separating the components physically or functionally

and then rearranging them differently.

• For instance, the first air conditioners came as a single unit. With the
division technique, later models separated the motor and fan from the
cooling unit, which minimized noise, heat and space requirements.

• Task Unification: Task unification emerges when innovators add a new

unrelated function to an existing component.

• An example is printing assembly instructions on the packaging of a

product. This eliminates the need for a separate sheet that customers
often lose, saves costs on printing the paper, and simplifies the
packaging process of the manufacturer.

• Attribute Dependency: Attribute dependency entails innovation arising

from beneficial connections between a product and its environment.

• For example, eyeglasses with lenses that change color when it is sunny
or dark eliminate the need to buy multiple pairs of eyeglasses.

Design Thinking

Empathize: To create meaningful innovations, you need to know your users

and care about their lives. Understand users’ lives, needs and emotions.

Define: Framing the right problem is the only way to create the right solution.
Craft POV and HMW. Define mode is sensemaking.

Ideate: It’s not about coming up with the ‘right’ idea, it’s about generating
the broadest range of possibilities. Brainstorm.
In the Define mode you determine the specific meaningful challenge to take
on, and in the Ideate mode you focus on generating solutions to address that
challenge.

Prototype: Build to think and test to learn. Keep low-cost. Visual Mental
Imagery.

Another ideation technique is building – that is, prototyping itself can be an

ideation technique. In physically making something you come to points
where decisions need to be made; this encourages new ideas to come
forward. There are other ideation techniques such as bodystorming,
mindmapping, and sketching. But one theme throughout all of them is
deferring judgment – that is, separating the generation of ideas from the
evaluation of ideas.

Test: Testing is an opportunity to learn about your solution and your user.
Show, don’t tell; create experiences; compare alternatives.

Guell Application:

Refrigerator and appliance industry.

While customers can be a great resource for innovation, they may not always
know what they need or lack the imagination to help develop innovations
(Goldenberg et al. 2003). As opposed to many other processes, systematic
inventive thinking is characterized by “inside the box” thinking and listening
to the voice of the product rather than solely to that of the consumer.

Ovens, dishwasher, refrigerator, stoves.

Customers want brands that provide entire product line.

Sentimental Analysis

• Common words (dryer, vent, lint) appear in both → polarity depends on

context

• Positive: drill associated with power & success; Negative: brush

associated with breakage

How TextBlob Computes Polarity

• Each word has a polarity score between -1 (negative) and +1 (positive)

• Sentence polarity = average of word scores (with some basic rules)

 • Example: 'This dryer kit is great but the brush broke quickly'

• → 'great' = +0.8, 'broke' = -0.6 → final polarity ≈ +0.1

Very few reviews are negative

On the left side, polarity < 0 shows only small bars.

This means:

Negative reviews are rare.

3️⃣ A small number of reviews are highly positive (0.6–1.0)

The right tail shows some bars for very positive reviews.
This means:

There are some extremely enthusiastic reviews, but not many.

4️⃣ The distribution is skewed right (positive)

Most bars are above 0.0, showing a general positive bias in the dataset.

• Overall sentiment appears mildly positive, not strongly enthusiastic

• Negative experiences may be underestimated → hidden in near-neutral

scores

• Lexicon method averages praise and complaints → masks mixed

reviews
 • Distribution shape shows perception is lukewarm rather than strongly
supportive

How the Co-Occurrence Network is Built

• Nodes = words from reviews

• Edges = words that appear together in reviews (pairs)

• Edge weight = frequency of co-occurrence across all reviews

• Graph shows not just which words are common, but how they relate to
each other

Aspect-Based Sentiment Analysis (ABSA)

ABSA breaks down sentiment by aspect (feature).

• SCAMPER technique: Substitute, Combine, Adapt, Modify, Put to

another use

Come up with entirely new product or improve the existing ones. Idea
generation. Add this in your analysis. Explain process of construction.

Come up with entirely new product or improve the existing ones. Idea
generation. Add this in your analysis. Explain process of construction.

Textblob polarity. Derived directly from the text of the reviews.

Make a small dictionary that maps some common game words like fun,
addictive, boring, ads, crash, puzzle, graphics, etc. to emotions like joy,
anger, fear, sadness, trust. Write a function that gives me the emotions for a
word.

Count emotions per review

For each review, count how many words belong to each emotion and save
the results.

Overall emotion totals

Add up all the emotion counts from all reviews and make a bar chart of the
results.

Aspect dictionary
Make a dictionary of aspects such as ads, gameplay, graphics, performance,
monetization, and privacy. Each aspect should have a few related keywords.
Tag aspects in reviews
Mark each review for whether it talks about these aspects based on the
keywords.

Aspect emotion counts

For each aspect, add up the emotions of the reviews that mention it and
show the top emotions.

## 11) Wrap-Up

- You extracted **emotion distributions** across reviews.

- You pinpointed **aspect-level** hotspots (e.g., ads → anger; gameplay →

joy).

- You turned evidence into **How Might We** prompts and **feature
ideas**.

- For more depth, swap in a full NRC lexicon and consider:

- weighting tokens by frequency (TF-IDF) per emotion,

- lemmatization to unify word forms,

- joining with **topic models** to discover hidden themes,

- comparing **before/after** app updates or across

**countries/segments**.

For lda.

Attributes then:

# Step 6: Compute sentiment scores for each perceptual attribute

# ---------------------------------------------------------------

# What this does:

# For each review, we look for words that belong to each perceptual
attribute

# (e.g., challenge, beauty, relaxation). If such words appear,

# we use the VADER sentiment analyzer to measure how positive or negative

# the review text is toward that attribute.

#

# The resulting columns (e.g., sent_challenge, sent_relaxation, etc.)

# store these sentiment scores on a scale from -1 (very negative) to +1

(very positive).

# If the attribute’s keywords are not found in a review, we mark it as NaN

(missing).

# Step 7: Aggregate average sentiment per attribute per game

# ------------------------------------------------------------

# What this does:

# For each game (e.g., BridgeMaster_f, Jigsaw_p, WordGame_f),

# we compute the average sentiment score for each attribute

# across all its reviews.

# This gives us a high-level view of how each game is perceived

# on challenge, relaxation, beauty, reward, value, usability, and variety.

# Step 8: Run PCA to reduce the 7 attribute sentiments into 2 dimensions

# ----------------------------------------------------------------------

# What this does:

# PCA (Principal Component Analysis) combines the seven sentiment

dimensions

# (challenge, relaxation, beauty, reward, value, usability, variety)

# into two new axes — "Dimension 1" and "Dimension 2".

# These two dimensions explain most of the variation across the games,

# helping us visualize how they differ perceptually.

# Step 9: Create a perceptual map (biplot)

# -----------------------------------------

# What this does:

# 1. Renames the sentiment columns to cleaner attribute names (e.g.,

"Challenge").
# 2. Plots each game based on its PCA scores (Dim1 and Dim2).

# 3. Draws attribute vectors (arrows) to show how each attribute influences

the map.

# Step 10: Add a preference (average rating) vector to the perceptual map

# -----------------------------------------------------------------------

# What this does:

# 1. Computes each game's average star rating and adds it as a new

variable called "Preference".

# 2. Standardizes this preference variable along with the seven perceptual

attributes.

# 3. Reruns PCA so that the preference direction is included in the same

space.

# 4. Plots a new biplot where the red arrows show perceptual attributes

# and a blue arrow shows the direction of increasing preference.

# Why this is useful:

# - It allows us to see which attributes align with higher user preference

(ratings).

# - Although not a formal joint-space map, it helps students understand

# how perceptions and overall liking can be visualized together.

Standardize the six columns in df_overall so that all dimensions are on the
same scale. This helps PCA treat every attribute equally.
10) Add a preference vector (optional but recommended): compute each game’s
average Rating and call it Preference. standardize it together with the 7
attributes, rerun PCA, and plot a new biplot where attribute arrows are red and
the Preference arrow is blue to show the direction of increasing ratings.
1. Group the data by Course and calculate the average for each of the six
perceptual dimensions. This gives one row per course showing the mean
perception across the four groups. Save this as df_overall.
2. 6️⃣ Standardize the Values
Standardize the six columns in df_overall so that all dimensions are on the same
scale. This helps PCA treat every attribute equally.

Mobile Game

Gap Analysis:

 Gap analysis in marketing refers to identifying unmet or underserved

needs in a market — the “white spaces” where customer expectations
are not fully satisfied by current offerings.

 Perceptual mapping provides a visual framework to locate those gaps

by showing how consumers perceive competing brands or products on
key attributes.

e.g. dualingo, emirates

Zach Case Study: Pricing

Zach’s Prompts

 Load the basic Python libraries for data handling, math, and plotting,
then read the Excel file named [Link] and show me the first few rows so
I can understand what the data contains.

 Convert the ratings in the six columns (d1, d3, d5, d8, d12, d20) into
purchase probabilities using this mapping: 5 becomes 1.0, 4 becomes
 0.4, 3 becomes 0.1, and both 1 and 2 become 0.0. After converting
them, show me the transformed dataset.

 Compute the average purchase probability at each of the six price

points. The prices are 1, 3, 5, 8, 12, and 20 dollars. Create a small
table that pairs each price with its average likelihood.

 Fit a logistic demand curve to the price and average-likelihood data.

Use this exact functional form:
likelihood = C / (1 + exp(-(alpha + beta * price)))
Estimate the parameters C, alpha, and beta using a nonlinear curve-
fitting approach.

 Generate predicted likelihoods over a smooth range of prices spanning

from the lowest price to the highest price. Use the fitted logistic model
to compute predicted purchase probabilities across this range.

 Plot the observed average purchase likelihoods as points and plot the
fitted logistic curve as a smooth line on the same graph, with price on
the horizontal axis and likelihood on the vertical axis.

 Estimate the expected revenue at each price along the smooth price
range. Use the formula:
expected revenue = 250 customers × 12 purchase opportunities per
year × predicted purchase probability × price.
Create a table or array of these revenue values.

 Plot the expected revenue against price so that I can see how revenue
changes as the price increases.

 Identify the price at which expected revenue is maximized. Report that

optimal price and the corresponding revenue value.

 Plot the revenue curve again and visually mark the optimal price by
adding a vertical line at that price, placing a point at the maximum
revenue, and adding a label with the revenue value.

Zach wanted to price it correctly for the concert. Didn’t want to charge from
the younger population so maximum people could attend his event.

Mark stood at the door of Zach’s Garage a few event nights, and asked
visitors how likely they would have come to that concert if they had to pay
an entrance fee equal to the following options: $1, $3, $5, $8, $12, and $20.
To reduce order effects, he randomized the order of the price levels
presented to a participant. Mark also tried to get a representative sample of
the target population, by randomly selecting only some of the event
attendees. He also asked a few demographic questions, such as age and
gender.
Look at complete pricing slide

Wan Vest Pricing

Testing:

LenDen

A/B testing
Expensive to collect oney and verification

Small amounts

InstaMoney Product A/B Testing — Summarized Notes

Overall Context

 LDC conducted multiple iterative A/B tests after launching the beta
version of the InstaMoney app.
  Goal: identify winning features at each step of the customer journey to
build the final product.

 Tests correspond to the customer journey stages: Lead → Prospect →

Qualified Borrower → Repayment → Customer Support.

1️⃣ Step 1: Lead → Prospect Conversion

 Definition

o Leads: users who showed interest in borrowing.

o Prospects: confirmed potential borrowers who paid a processing

fee of ₹199.

 Experiment

o Tested website background colour aesthetics for borrowers

searching the loan app.

o Two color sets randomly assigned:

 Set A: green, blue, white (original LDC brand colours)

 Set B: purple, green, blue, white (new InstaMoney

scheme)

 Purpose

o Identify which colour scheme improves conversion from lead →

prospect.

2️⃣ Step 2: Prospect Qualification

 Prospects paying the ₹199 fee were verified for creditworthiness.

 Verification required government-approved documents such as:

o bank account details

o income tax return filings

 Experiment

o Two document submission methods:

1. Manual upload by the user

 2. Auto-fetch authorization allowing LDC to retrieve
documents from original sources (e.g., bank)

 Purpose

o Determine which method yields better prospect qualification

rates.

3️⃣ Step 3: Borrower Selection (Creditworthiness Assessment)

 LDC assessed borrowers using AI + ML credit algorithms

incorporating 120+ parameters.

 Soft Information examples:

o demographics, education, work history

o residence and past residence patterns

 Hard Information examples:

o income, credit history, repayment behaviour

o utility bill payment history, spending patterns

 Additional features included employer reputation, found to strongly

predict default behavior.

 Experiment

o Two weighting models:

 Model A: greater weight on location of residence

 Model B: greater weight on income

 Purpose

o Improve prediction accuracy for borrower default risk.

4️⃣ Step 4: Repayment Behaviour

 Defaults classified as:

o Intentional

o Unintentional (borrowers forgot due dates)

  Unintentional defaults reduced significantly when borrowers were
reminded in advance.

 Experiment

o A/B test for repayment reminders:

 Group A: voice bot reminders

 Group B: email/text reminders

 Purpose

o Identify which reminder format reduces default rates most

effectively.

5️⃣ Step 5: Customer Support & Query Resolution

 Borrowers contacted LDC for:

o ongoing loan queries

o new loan queries

 Quick and satisfactory responses were crucial for repeat business.

 Experiment

o A/B test comparing:

 Group A: email-based support

 Group B: chatbot-based support

 Performance metric

o Query resolution turnaround time, including follow-up to

confirm borrower satisfaction.

 Purpose

o Determine which support channel speeds up resolution and

improves user satisfaction.

📚 Final Summary
LDC performed A/B testing at every stage of the customer lifecycle to
optimize:

1. Website aesthetics

2. Document submission methods

3. Creditworthiness scoring weights

4. Repayment reminder channels

5. Customer support channels

Key components of concept testing

• Sample Selection

• Who should evaluate the concept?

• Should represent the target market or potential early adopters.

• Sample size and segmentation affect the reliability of insights.

• Concept Communication

• How is the product idea presented?

• Options: written descriptions, images, storyboards, videos, or

prototypes.
 • Clarity and realism are crucial for valid feedback.

• Consumer Response Measured

• What are we trying to learn?

• Measures include purchase intent, uniqueness, relevance, and

clarity.

• Can also capture open-ended feedback and emotional reactions.

Storyboarding for Concept Communication

 Storyboarding is a visual tool that illustrates how a new product or

service fits into a consumer's life.

It helps communicate usage context, emotional impact, and customer

journey in a simple, relatable way.

Best suited for:

• Service innovations
• Complex usage scenarios
• Early-stage concept feedback

 Physical Mock-ups or Prototypes: Tangible, often non-functional models

to test packaging or usability. E.g., a shampoo bottle prototype.

 Animatics / Storytelling Videos: Low-cost animations or narrated visuals

to simulate an ad or user scenario.

 Crowdsourced Platforms: Test concepts through platforms like

Kickstarter to gauge market interest and pre-orders.
  Simulated Shopping / Virtual Shelf Tests: Place product in a virtual store
to test consumer choice among competitors.

 Landing Pages & A/B Testing: Create variations of product pages to

track engagement or conversions.

 Purchase Intention Measures: How likely the respondent is to buy the

product (e.g., 5-point or 11-point scales).

 Overall Product Diagnostics: General feedback on the concept's appeal,

clarity, uniqueness, and believability.

 Specific Attribute Diagnostics: Ratings or open-ended responses on

specific features (e.g., taste, packaging, benefits).

 JAR (Just About Right) Measures: Evaluate whether an attribute is too

much, too little, or just right (e.g., sweetness, size).

 Emotion Metrics: Capture emotional reactions to the concept using

words or emoji scales (e.g., Excited, Confused).

 Respondent Profiling Variables: Collect demographic and behavioral

info: age, gender, usage frequency, lifestyle segment.

Purchase Intention Measures

 Based on this product description, how likely are you to buy this
product if it were available in a store near you? Please select one
response below.

 Definitely would buy

 Probably would buy

 Might or might not buy

 Probably would not buy

 Definitely would not buy

 For nondurable goods, the frequency of purchase is also key.

 In cases where the product may come in different sizes or multiple

units might be purchased at one time, these issues are also addressed.

 Sales volume per household in a time period = % households in market

who try × expected # purchases in the period for triers × expected #
units per purchase
Overall Product Diagnostics

With respect to overall product judgements, ask questions to address the

concept’s:

1. Uniqueness or differentiation from other products

2. Believability

3. Importance in solving a consumer's problem

4. Inherent interest

5. Value for the money

What the JAR Curve Shows

The JAR framework is used in product development (especially food,

cosmetics, beverages, UX, and physical products) to understand how
changing an attribute affects overall liking.

Examples of attributes:

 Sweetness

 Saltiness

 Spiciness

 Fragrance intensity

 Thickness / creaminess

 Brightness / contrast

 Size / weight
JAR tells you how far a product is from the consumer’s ideal level of
an attribute.

📈 Understanding the Curve

1️⃣ X-axis: Sensory Evaluation of an Attribute

This axis represents how consumers perceive the level of an attribute:

 Much too little

 Too little

 JAR (Just-About-Right)

 Too much

 Much too much

The JAR point is the optimum level—where most consumers feel the product
is “just right.”

2️⃣ Y-axis: Overall Liking

This is the consumer’s overall preference or satisfaction with the product.

 Highest liking at the peak

 Lowest liking at the extremes

✨ Key Insight: The Curve Is an Inverted U-Shape

 Peak = Optimum (JAR)

o This is the level where the attribute delivers maximum liking.

 Both sides of the curve show penalties:

o Too little → liking drops

o Too much → liking also drops

The further the attribute deviates from the JAR point, the bigger the penalty
to overall liking.
⚠️Penalty Concept

A penalty is the drop in consumer satisfaction when an attribute level is not

ideal.

 If sweetness is too low, consumers feel dissatisfied.

 If sweetness is too high, they also feel dissatisfied.

JAR analysis quantifies these drops and tells you:

✔ Whether you should increase or decrease an attribute

✔ How much improvement in liking you can gain by fixing the issue
✔ Which attribute deviations have the largest penalties (highest priority
for reformulation)

🧠 Why JAR Is Important in Product Development

JAR helps answer:

 “Is this product too salty or not salty enough?”

 “Is the fragrance intensity optimal?”

 “Is the product too thick or too thin?”

This turns vague consumer feedback into quantitative guidance for

product reformulation.

Companies use JAR to:

 Improve product formulas

 Fine-tune sensory attributes

 Optimize taste, texture, usability, fragrance, etc.

 Identify which changes will give the biggest boost to satisfaction

✔ Instructor Summary

The JAR measure is a diagnostic tool that shows how consumers’

perception of an attribute impacts overall liking.

 The goal is to hit the JAR zone at the top of the curve.

 Deviations on either side produce penalties that lower satisfaction.

  The tool helps developers understand whether an attribute needs to be
increased or decreased to get closer to the optimum.

Key Profiling Variables

1. Demographics

 Age, gender, income, education, household size, etc.

 Useful for identifying broad consumer segments.

2. Psychographics

 Lifestyle, attitudes, motivations, values, interests.

 Helps interpret why consumers behave a certain way.

3. Behavioral and Perceptual Variables

These go beyond who consumers are, focusing on how they think and act.

a. Current purchase behavior

 What they buy today

 Purchase frequency and volume

 Typical spend levels

b. Perception of the category

 How they view the product category (e.g., essential, luxury, problem-
solving, risky, etc.)

 Whether they see differentiation between brands

c. Barriers to switching brands

 Habit or loyalty

 Convenience

 Price sensitivity

 Trust or perceived risk

 Lack of awareness of alternatives

d. Influence in the purchase decision

 Whether the respondent is a primary decision-maker

 Whether they share or delegate purchasing responsibility

 Strength of influence on household or group buying

Example Insight

 Consumers with high purchase intent might still stay with their
current brand if they also report:

o High satisfaction with their existing brand

o Strong loyalty or switching barriers

 Meaning: Even high interest in a new product does not guarantee

switching if current satisfaction is strong.
Review testing concept slide again if possible

Exercise: TURF Analysis for Snack Flavor Selection

A company wants to launch a new line of snacks and is considering offering

three flavors: Barbecue, Sour Cream & Onion, and Cheese. They conducted a
survey with 10 participants, asking which flavor they would buy. The results
are shown below:

Participan Barbecu
Sour Cream & Onion Cheese
t e

1 Yes No Yes

2 No Yes No

3 Yes Yes No

4 No Yes Yes

5 Yes No No

6 No No Yes

7 Yes Yes No

8 No Yes Yes
 Participan Barbecu
Sour Cream & Onion Cheese
t e

9 Yes No Yes

10 No Yes Yes

1. Calculate the reach for each of the three flavors individually (i.e., how
many participants said "Yes" to each flavor).

2. Then, calculate the unduplicated reach for each possible two-flavor

combination:

o Barbecue + Sour Cream & Onion

o Barbecue + Cheese

o Sour Cream & Onion + Cheese

3. Based on your calculations, which two-flavor combination offers the

largest total reach (i.e., the highest number of participants reached)?

4. Frequency refers to how many flavors each participant would buy.

o Calculate the average frequency for each two-flavor

combination.

Steps for Calculation:

 Reach of each flavor: Count how many participants said "Yes" to each
flavor.

 Unduplicated reach of two flavors: Count the number of unique

participants who would buy either of the two selected flavors (without
duplication).

 Frequency: Add up the total number of "Yes" responses for both flavors
in each combination and divide by the number of participants reached.

Answers:

Task 1: Total Reach (Two-Flavor Combinations)

Reach of Each Flavor:

 Barbecue:

o Participants 1, 3, 5, 7, 9 said "Yes".

o Total: 5 participants

 Sour Cream & Onion:

o Participants 2, 3, 4, 7, 8, 10 said "Yes".

o Total: 6 participants

 Cheese:

o Participants 1, 4, 6, 8, 9, 10 said "Yes".

o Total: 6 participants

Unduplicated Reach for Two-Flavor Combinations:

 Barbecue + Sour Cream & Onion:

o Participants 1, 2, 3, 4, 5, 7, 8, 9, 10.

o Total unduplicated reach: 9 participants

 Barbecue + Cheese:

o Participants 1, 3, 4, 5, 6, 7, 8, 9, 10.

o Total unduplicated reach: 9 participants

 Sour Cream & Onion + Cheese:

o Participants 1, 2, 3, 4, 6, 7, 8, 9, 10.

o Total unduplicated reach: 9 participants

Task 2: Reach and Frequency (Two-Flavor Combinations)

Average Frequency for Two-Flavor Combinations:

 Barbecue + Sour Cream & Onion:

o Total "Yes" responses = (5 for Barbecue + 6 for Sour Cream &

Onion) = 11.

o Average frequency = 11 / 9 participants = 1.22

 Barbecue + Cheese:
 o Total "Yes" responses = (5 for Barbecue + 6 for Cheese) = 11.

o Average frequency = 11 / 9 participants = 1.22

 Sour Cream & Onion + Cheese:

o Total "Yes" responses = (6 for Sour Cream & Onion + 6 for

Cheese) = 12.

o Average frequency = 12 / 9 participants = 1.33

Combining reach and frequency is a practical approach to resolving ties in

TURF analysis. This method ensures that you maximize the number of
customers reached while also considering how many flavors each customer
is likely to purchase (frequency). If two combinations have the same reach,
you can prioritize the one with the higher frequency to increase sales
volume.

So, which combination would you make?

Bass Model review it

I’ll walk through the note like I’m teaching you the Bass model from scratch,
and I’ll hit every key concept that appears in the document.

1. What the Bass Model Is For

Goal:
Predict how a new product or technology will be adopted over time by a
population of potential customers, when there are no close competitors.

It answers two big questions:

1. How many people will eventually adopt?

2. When will they adopt? (shape of the sales curve over time)

This is crucial because the timing guides marketing spend, production

capacity, launch strategy, etc.

The model is built around a contagion idea: people adopt because of:

 their own inherent tendency to try new things, and

 influence from others who have already adopted (word-of-mouth).

2. The Probability Functions: F(t), f(t), and L(t)

We start from a probability view of a single customer.

2.1 F(t): Cumulative probability of adoption

 (F(t)) = probability that a person has adopted by time (t).

 It’s non-decreasing and approaches 1 as time → ∞ (eventually

everyone in the target group adopts).

 In the graph on page 3, top panel (Exhibit 1a), F(t) is the S-shaped
curve: low at the start, rising, then flattening.

2.2 f(t): Probability density (instantaneous rate)

 (f(t) = \dfrac{dF(t)}{dt})
  It tells you how fast that probability is changing at time t – i.e., the
density of adoption timing.

 In the bottom panel of Exhibit 1, f(t) is the bell-shaped curve:

adoptions per unit time start low, peak, then decline.

You can think of F(t) as “% of population who have adopted by time t,” and
f(t) as “% who adopt at exactly time t.”

2.3 L(t): Conditional likelihood (hazard rate)

The note defines:

[
L(t) = \frac{f(t)}{1 - F(t)}
]

This is the conditional likelihood that a person adopts at time t, given

that they haven’t adopted before t.

Name for this in statistics: the hazard rate.

Intuition:

Among all people still sitting on the fence at time t, what fraction adopts
right now?

This is what Bass models directly.

3. The Core Bass Idea: Innovation and Imitation

Bass proposes a simple structure for (L(t)):

[
L(t) = p + q \frac{N(t)}{N}
]

Where:

 N(t): number of people who have already adopted by time t

 N: total market potential (size of the target segment that will

eventually adopt)

 p: coefficient of innovation (external influence)

 q: coefficient of imitation (internal influence)

3.1 What p and q mean

p (innovation coefficient)

 Captures adoption that happens independently of others.

 Influences: mass media, advertising, promotions, or just people who

like trying new things (“innovators”).

 It is a constant baseline probability of adoption per unit time.

q (imitation coefficient)

 Captures adoption that happens because others already adopted.

 Driven by word-of-mouth, social proof, observing others using the

product.

 The term (q \frac{N(t)}{N}) grows as more people adopt, so imitation

effects strengthen over time, then eventually weaken when few
non-adopters remain.

So, L(t) = p + imitation-effect.

4. From Hazard Probability to Sales Equation

Using relationships among F(t), N(t), and n(t):

 (N(t) = N F(t)) (cumulative adopters = total market × cumulative

probability)

 (n(t)): number of new adopters at time t (sales at time t)

After some algebra, the note derives:

[
n(t) = \left(p + q \frac{N(t)}{N}\right)\left[N - N(t)\right]
]

or commonly written as:

[
n(t) = \left(p + q \frac{N(t)}{N}\right)\big[N - N(t)\big]
]

Same as Eq. (4) in the note (just rearranged).

Interpretation:
  (N - N(t)) = remaining potential (people who haven’t yet adopted).

 The term in parentheses is the probability of adoption in this

period.

 So sales at time t = “how many people are left” × “probability each

will adopt now”.

4.1 Shapes of the sales curve

Depending on relative sizes of p and q:

 If q > p (imitation dominates)

o Sales curve (n(t)) is inverted U-shaped (classic S-curve

diffusion).

o Slow start → rapid growth → decline.

 If q < p (innovation dominates)

o Most adoption happens early; sales peak at launch and

decline afterwards.

o Example: blockbuster movies with huge opening weekends.

 Low p: diffusion starts slowly; takes time for sales to ramp up.

 High p and high q: fast takeoff and fast decline (steep life cycle).

By adjusting p and q you can mimic many real-world adoption patterns.

5. Generalized Bass Model: Adding Marketing Mix

Basic Bass ignores explicit marketing actions. The Generalized Bass model
adds a term for marketing effort:

[
f(t) = \big[p + q\frac{N(t)}{N}\big],[1 - F(t)] , x(t)
]

where x(t) represents the effect of marketing mix variables in period t

—typically price and advertising.

The suggested form for x(t) is:

[
x(t) = 1 + \alpha \bigg[\frac{P(t-1) - P(t)}{P(t-1)}\bigg]
+ \beta \bigg[\frac{A(t) - A(t-1)}{A(t-1)}\bigg]
]

(That’s what Eq. (6) is describing in words.)

Conceptually:

 P(t): price at time t

 A(t): advertising at time t

 α: % increase in diffusion speed from 1% price decrease

 β: % increase in diffusion speed from 1% advertising increase

So:

 Lower price → x(t) > 1 → diffusion speeds up.

 Higher advertising → x(t) > 1 → diffusion speeds up.

We measure marketing effort relative to a base level (indexed to 1).

If we double the base advertising level, x(t) might be 2.0, meaning the
effective adoption rate doubles.

Takeaway: marketing effort multiplies the baseline Bass diffusion

process rather than replacing it.

6. Estimating the Parameters (N, p, q, α, β)

There are two broad approaches:

6.1 Using historical sales data (data-based methods)

If we have several periods of sales data for the product:

 Use nonlinear regression (nonlinear least squares).

o Discretize the model: express sales in period t as a function of

cumulative sales up to t–1.

o Example discrete form (Eq. 7):

[
n(t) = \big[p + q \frac{N(t-1)}{N}\big],[N - N(t-1)]
]
 o Fit p, q, and N to minimize squared error between predicted and
actual sales.

Other, more advanced estimation methods:

 Maximum likelihood estimation (MLE)

 Hierarchical Bayes (HB) – especially when you have multiple

products/segments.

These advanced methods require knowing launch date and more detailed
data.

6.2 Using judgment / analogs (judgmental methods)

When you don’t have much sales history (e.g., pre-launch):

 Use analogs: find similar products whose diffusion pattern is known

and borrow their p and q.

 Use expert judgment or surveys of long-term purchase

intentions to estimate market potential N.

Practical recommendation from the note:

 Try to estimate N externally (e.g., long-term intentions survey).

 Then fit p and q using nonlinear regression on early sales.

 For the Generalized Bass, estimate p and q as usual, and let

marketing managers specify α, β based on their judgment (since
mapping marketing to diffusion speed can be fuzzy).

7. Using the Bass Model for Forecasting

Once you have N, p, q (and maybe α, β):

1. Put them into a spreadsheet.

2. For each period:

o Compute sales (n(t)) using the Bass formula.

o Update cumulative sales (N(t)) = previous cumulative + current

sales.

3. This gives you:

 o Sales per period (new adopters).

o Cumulative adoptions over time.

On page 6 (Exhibit 2) they show an example:

 Product: room-temperature control unit.

 Parameters: p = 0.01, q = 0.41, N = 16,000 (in thousands).

 The table lists quarterly sales and cumulative sales.

 Sales start at 160 (Q1), rise to a peak (around Q8–Q12), then decline;
cumulative sales approach 16,000.

Key point: the software or spreadsheet basically automates these repeated

calculations.

8. Limitations of the Basic Bass Model

The note warns about important limitations:

1. Success bias

o Most historical diffusion data come from successful products.

o If you use those analogs blindly, you may over-forecast

because failures are under-represented.

2. Late parameter estimation

o You estimate p and q after several periods of sales, but by

then major investment decisions are already made.

3. Choice of analogs

o Pre-launch forecasting relies on picking the “right” analogous

product; this requires judgment and can be wrong.

9. Extensions: Relaxing the Bass Assumptions

The basic model makes several simplifying assumptions. More advanced

models relax them:

1. Fixed market potential N

 o In real life, the target market size can grow (population growth,
new segments, lower uncertainty, price drops).

o Some models let N vary over time (e.g., with segment growth
or technology uncertainty).

2. No marketing influence

o Addressed by the Generalized Bass model, which includes

price, advertising, selling effort, etc.

3. Binary decision: adopt vs. not adopt

o Reality: multi-stage process: awareness → interest → trial →

adoption → word-of-mouth.

o Multi-stage diffusion models handle these transitions.

4. Constant imitation coefficient q

o Word-of-mouth strength may vary over the life cycle (strong early
on, then weaker).

o Some models allow q(t) to change over time.

5. Uniform mixing

o Basic Bass assumes everyone can influence everyone else

equally.

o Network-based models use social networks / small-world

structures with close and distant ties.

6. Imitation is always positive

o Reality: word-of-mouth can be negative too.

o Models with both positive and negative WOM show different

optimal marketing strategies:

 Positive WOM (“sleeper” movie): ramp up marketing

slowly, let WOM build.

 Negative WOM risk (“mega-bomb” movie): heavy early

marketing to get quick trials before bad news spreads.

7. Independence from other innovations

 o Some products depend on related technologies (e.g., multimedia
software needs powerful PCs).

o Complementarity models jointly model adoption of

interdependent products.

8. No repeat or replacement purchases

o Basic Bass only models first-time adoption.

o Extended models add repeat and replacement purchases to

forecast total sales, not just first adoptions.

10. Big Picture Summary

 The Bass model is a diffusion model describing how a new product

spreads through a population.

 It is built on:

o Market potential N

o Innovation effect p (external influence)

o Imitation effect q (internal influence)

 It provides:

o A closed-form, S-curve adoption pattern

o Intuitive interpretation (innovators vs. imitators)

o A way to forecast long-term sales using a few parameters.

Over time, researchers have built many variations that add reality: marketing
mix, networks, multi-stage adoption, complements, replacement, etc.

If you want, next step we can:

 Take a real dataset (or a small hypothetical one) and

 Fit a Bass model step-by-step in Excel or Python,

 Plot the diffusion curve, and interpret p and q in practical, marketing

terms.

Durr
SUMMARY — Dürr Environmental, Inc. Case (Marketing Engineering
for Excel)

Purpose of the Case

Dürr Environmental, a German firm with strong service networks in the U.S.,
is evaluating whether to enter the U.S. market with a new air pollution
control system.
They must determine:

1. Whether entry is economically justified.

2. The best product offering(s).

3. Whether one or two offerings should be launched.

4. Which customer segment(s) to target and with what value proposition.

The business must generate $4M/year in marginal revenue to justify entry;

a second offering adds $3M/year in additional cost.

Industry Background

 Air pollution control in the U.S. is driven by regulatory pressure (e.g.,

Clean Air Act).

 Key pollutants targeted:

o VOCs (Volatile Organic Compounds) → contribute to

ozone/smog

o NOx (Nitrogen Oxides) → contribute to smog and acid rain

 A complete air emission control system performs:

1. Filtering (removing particulates)

2. Adsorption (concentrating particulates)

3. Oxidation (converting pollutants to CO₂ & water)

 Dürr’s focus: thermal oxidizer system (50,000 SCFM capacity).

 Market size: ~300 units/year over next decade (U.S.).

Customer Purchase Drivers (Key Attributes)

Customers evaluate systems based on four main factors:

1. Efficiency

o % of heat recycled

o EPA target = 90%; some customers want higher for compliance

or corporate citizenship.

2. Delivery Time

o Months from order to installation.

3. Price

o List price for 50,000 SCFM system.

4. Delivery Terms

o Installed vs. skid-mounted

o Warranty/service contract length (industry standard = 60 days)

Competitive Landscape

Competito Delivery Pric

Efficiency Delivery Terms
r Time e

Waste 5% above $600

9 months FOB + Service contract
Watch EPA k

Thermatri 9% above $900

12 months FOB + Service contract
x EPA k

Advanced $600 Installed + Service

Meets EPA 9 months
Air k contract

Market shares:

 Waste Watch: 45%

 Thermatrix: 5%

 Advanced Air: 50%

Competitor weaknesses:

 Waste Watch & Thermatrix lack a strong service network.

  Advanced Air has strong service but dated technology.

Dürr’s Advantages & Uncertainties

Advantages

 Strong U.S.-based service representatives.

 Belief that Dürr systems require less service than competitors.

 Engineers believe they can offer up to 2-year warranty (vs. 60 days

norm).

 New technology may allow up to 9% above EPA spec, exceeding

current market capabilities.

Uncertainties

 How much customers value a 2-year warranty.

 Optimal product configuration for entry.

 Which segments value efficiency, service, speed, or price most.

Market Research Conducted

1. Conjoint Analysis (Preference Measurement)

 Data in Durr Data (Conjoint).xls.

 31 companies participated (sampled across food, energy, and other

industries).

 Customers evaluated bundles varying along efficiency, delivery

time, price, and delivery terms.

 Output:

o Part-worth utilities per firm

o Attribute importances

o Enables market share predictions using the Maximum Utility

rule.

2. Segmentation Study
  Data in Durr Data (Segmentation).xls.

 Uses conjoint outputs + firm-level descriptors:

o Industry

o Sales, profit %, ROE

o Employees

o Growth rate

o Strategic priorities (100-point allocation across Growth, Profit,

Market Share, Tech Leadership, Regulation, Citizenship)

o Influence weights by job function (Management, Engineering,

Finance, Purchasing)

 Goal: identify distinct customer segments with different preference

patterns and strategic motives.

Product Design Options

Dürr engineers propose three possible entry products:

1. Servair DX

 Meets EPA specs

 15-month delivery

 $900k

 Installed + 2-year warranty

 Higher warranty cost (~$150k/year to support)

2. Premier LX

 Exceeds specs by 9%

 12-month delivery

 $900k

 Installed + service agreement

 Targets Thermatrix’s segment (premium tech but weak service)

3. Base Product
  Exceeds specs by 5%

 12-month delivery

 $700k

 Installed + service agreement

 Basic entry model

Cost Structure for Product Components

Base product cost = $500k

Selling price = $700k → $200k/unit margin

Marginal cost adjustments apply for changes in price, efficiency, delivery

time, and terms.

Examples:

Price

Pric Cost
e Adj.

$600
−$100k
k

$700
0
k

$800
+$100k
k

$900
+$200k
k

Efficiency

Cost
Level
Adj.

Exceed
−$170k
9%
 Cost
Level
Adj.

Exceed
0
5%

Meet
+$70k
specs

Short by
+$90k
5%

Delivery Time

Month Cost
s Adj.

6 −$40k

9 −$20k

12 0

15 +$10k

Delivery Terms

Installation cost ≈ $100k

Service contracts break even for Dürr.

Cost
Term
Adj.

Installed, 2-year
+$10k
warranty

Installed, 1-year
−$300k
warranty

Installed, service
−$100k
contract

FOB + service
+$100k
contract

Total gross margin = Base margin ($200k) + adjustments.

Example (Servair DX):
$200k + $200k (price) + $10k (delivery time) + $70k (efficiency) − $300k
(terms) = $180k margin

Analytical Tasks Required

1. One-Product Strategy Analysis

Use all respondents (weighted) to calculate:

 Market share using conjoint analysis

 Expected annual profit =

(300 units)×(market share)×(unit margin)−$ 4 M development cost

2. Two-Product Strategy Analysis

Steps:

1. Segment customers using ME►XL Segmentation and Targeting

tool.

2. Identify distinct clusters and their preferences.

3. Assign a product to each segment.

4. Compute profit:

∑ (segment size ×product share × margin)−(market dev. costs +$ 3 M extra)

segments

3. Recommendations

Answer:

 Should Dürr enter the U.S. market?

 Which product(s) maximize margin and market share?

 Which segments should be targeted?

 What is the best positioning strategy?

Exhibits Provided

 Conjoint bundles used in survey

  Raw respondent ratings

 Part-worth utilities for each firm

 Segmentation variables (strategic goals, decision influence, firm

metrics)

These are used to calculate:

 Preferences

 Market shares

 Segment attractiveness

 Profitability of each launch strategy

FINAL SUMMARY

Dürr Environmental is evaluating entry into the U.S. VOC/NOx pollution

control market with a new thermal oxidizer system.

They face strategic decisions about:

1. Whether to enter

2. What product configuration(s) to launch

3. Which customer segments to target

4. How to position vs. incumbents

The case requires:

 Conjoint analysis to measure customer preferences

 Segmentation analysis to identify customer clusters

 Cost and profit modeling to evaluate each possible product offering

 A business case determining whether one or two products are optimal.

Using Marketing Engineering for Excel, students must estimate market

shares, margins, segment attractiveness, and profitability to recommend a
launch strategy for Dürr.

Forte Hotel Design — Summary

Forte Hotels, the UK’s largest hotel chain, is planning to launch a new U.S.
business-hotel brand called Forte Executive Innes. The goal is to blend
European ambiance with American business functionality, appealing
especially to rising numbers of European business travellers visiting the
U.S., while also attracting American business guests seeking comfort and
service.

Market Insights

 Business travellers choose hotels mainly based on price, location,

and brand name.

 Men emphasize price and convenience; women prioritize

cleanliness and safety.

 Repeat visits depend more on specific hotel attributes than on price.

 Desired amenities among at least 30% of travellers include:

o In-room computer/Web access

o On-site conference rooms

o Large desks/workspaces

o Speakerphones, data ports, fax machines

 European travellers split between:

o Those wanting comfort and relaxation

o Those wanting efficiency to finish work quickly

To design a compelling offering, Forte needs a deep understanding of

customer preferences.

Conjoint Analysis Approach

Forte identifies five key hotel attributes, each with multiple design options
(Exhibit 1, p. 4):

1. Room Type

2. Business Amenities (Web access, speakerphone, in-room fax)

3. Leisure Facilities (exercise room, pool, combination)

 4. Conveniences & Extras (shoe shine, video library, fruit & cheese,
newspaper)

5. Restaurant Delivery Service

Because attributes are mix-and-match, the firm cannot intuitively know

which bundle customers value most.
Thus, they conduct a conjoint analysis using evaluations from 300
travellers, with a working dataset of 40 respondents for the classroom
exercise (Exhibit 2, pp. 5–6).

Conjoint analysis will help Forte:

 Quantify part-worth utilities for each attribute level

 Determine attribute importance

 Predict customer preference for full hotel concepts

 Evaluate the market viability and profitability of different hotel

designs

Case Tasks (Exercises)

Students use ME►XL Conjoint tools to:

1. Build a study design template and data collection instrument for

the chosen attributes.

2. Generate utility estimates (part-worths) for each option.

3. Understand advantages/limitations of conjoint for preference

measurement.

4. Analyze four proposed hotel concepts using the provided 40-

respondent dataset and the cost data (Exhibit 3, p. 6).

5. Identify the optimal hotel concept for the State College market.

6. Optionally propose alternative designs not considered by Forte.

7. Evaluate when conjoint analysis is appropriate in the hotel industry.

8. Discuss whether conjoint analysis stifles creativity—a debate prompt

about balancing data-driven design vs. visionary innovation.
Cost and Profit Considerations

Exhibit 3 (p. 6) provides incremental cost per room and expected

incremental daily revenue for each optional feature:

 Some features generate positive returns (e.g., Web access,

speakerphone, fax).

 Others reduce expected revenue despite higher cost (e.g., pool, fruit &
cheese bowl).

 Base room construction cost (without options): $40,000 per room for
150–200 rooms.

These financial data allow evaluation of whether a given hotel concept

creates sufficient revenue to justify its additional cost.

Overall Case Purpose

The Forte case teaches students how to:

 Use conjoint analysis to guide new product/service design

 Make attribute trade-off decisions grounded in customer

preference data

 Evaluate market viability and financial performance

 Understand limitations of data-driven design and the role of

managerial judgment

Forte’s strategic challenge is to choose the most appealing and profitable

combination of hotel attributes for its new U.S. chain, supported by analysis
rather than intuition.

Hotel Concept Simulations – Outline

1. Use the ratings data from respondents to understand how much value
they place on different hotel features such as room type, business amenities,
leisure facilities, extras, and restaurant delivery.

2. Run regression models for each respondent to estimate their part-worth

utilities, which capture how each attribute level contributes to their overall
preference.
3. Adjust and scale these utilities so they can be compared across
respondents and attributes. This ensures that everyone’s preferences are
measured on a common scale.

4. Apply these estimated utilities to the profiles of all hotel concepts, both
existing and new, to compute the total preference score for each concept.

5. Simulate market behavior using the first-choice rule, where each

respondent “chooses” the hotel concept that gives them the highest total
utility.

6. Compare baseline market shares among existing hotels, and then add
each new concept one at a time to see how it changes the market—revealing
which new concept is likely to attract the most customers.

Prompts for the Hotel Conjoint Workflow

1) Load ratings data

- Load “/content/forte_ratings.csv” into a DataFrame named forte.
- Confirm dtypes; ensure Rating is numeric and the attribute columns (Room,
BusAmenities, Leisure, Extras, RestDelivery) are treated as categorical.

2) Quick structure check

- Show the first 5 rows and basic .info() to verify columns and missing values.

3) Pooled (all-respondents) regression model

- Create dummy variables for all attribute columns. Use full dummies and no
intercept so part-worths are directly interpretable per level.
- Fit OLS with Rating ~ 0 + dummies on the entire dataset to get pooled part-
worths. Save coefficients as pooled_betas.

4) Single-respondent model (Amanda)

- Subset forte to Respondents == "Amanda".
- Using the same dummy specification, fit Rating ~ 0 + dummies for Amanda
only. Save as amanda_betas.

5) Iterate over all respondents

- Group by Respondents and fit Rating ~ 0 + dummies per respondent.
- Store each coefficient vector in a dictionary respondent_models.
- Convert to a DataFrame beta_matrix with respondents as rows and dummy
features as columns; display it.

6) Within-attribute zero-base scaling

- For each respondent and each attribute group (Room, BusAmenities,
Leisure, Extras, RestDelivery), subtract the minimum beta in that group so
each group’s lowest level is 0. Call this beta_matrix1.

7) 0–100 utility scaling for comparability

- For each respondent, sum the highest level within each attribute across the
five attributes; call this sum denom.
- Divide all betas by denom and multiply by 100. Save as beta_matrix2.

8) Load the 9 hotel concepts

- Load “/content/hotel_concepts.xlsx” into hotel_concepts.
- Confirm the shape; display first 5 rows and the column names.
- Transpose so each row = concept and each column = attribute, with an
index or column holding concept names. Save as hotel_concepts_transposed.

9) Map concept levels to dummy names

- Build attribute_mapping: a dict from each raw level label in
hotel_concepts_transposed (e.g., “Large Room”, “Exercise room + Pool”) to
the matching dummy column in beta_matrix2.
- (Tip: create a helper that, for each attribute column, looks up the correct
dummy name pattern used during one-hot encoding.)

10) Align names and dummy-code the concepts

- Option A (rename betas): rename columns of beta_matrix2 using
attribute_mapping if necessary to ensure perfect alignment. Save as
beta_matrix2_renamed.
- Option B (preferred): build a concept design matrix
hotel_concepts_dummies whose columns exactly match beta_matrix2’s
dummy columns (fill missing with 0). Display first 5 rows.

11) Compute respondent-by-concept utilities

- Compute hotel_utilities = beta_matrix2 @ hotel_concepts_dummies.T (or
vice-versa, consistent with your shapes).
- Set column names of hotel_utilities to the concept names from
hotel_concepts_transposed and display the first few rows.

12) Baseline: incumbents only

- The first five hotels (Courtyard by Marriott, Nittany Lion Inn, Atherton Hilton,
Toftrees, Scanticon) represent the existing market.
- Compute each respondent’s utilities and determine their most preferred
hotel among these five.
- Calculate the market share (percentage of respondents choosing each
hotel).
- This serves as the baseline for comparing new concepts.

13) Add each new concept one at a time

- Add one new concept (Professional 1, Professional 2, Tourist, or Deluxe) to
the five incumbents to form a six-hotel market.
- For each respondent, find which of the six hotels gives the highest utility.
- Calculate the new shares for all six hotels.
- Record the share of the new concept and how much the incumbents’ shares
change.
- Repeat for each of the four new concepts to see which performs best
against the existing market.
Market Entry Decision for Durr’s New Air Pollution Control System

Durr is developing a new air pollution control system to compete against

existing offerings from Wastewatch, Thermatrix, and Advanced Air.
To guide its launch decision, Durr conducted a conjoint study with 31
business buyers in its target market. The study measured preferences over
key attributes such as:

 Thermal Efficiency

 Delivery Time

 Price
  Delivery Terms

Each buyer’s part-worth utilities for these attributes were estimated, and
a weight proportional to their purchase volume was assigned to reflect their
market importance.

Using these utilities, Durr simulated the perceived attractiveness of the

existing offerings and three new concepts under consideration:

 Servair

 Premier

 Base

Part 1: Evaluating the Single-Product Entry

Durr must first decide which single product to introduce.

Your task is to compute weighted market shares for Servair, Premier, and
Base using the provided Excel file.

For each product:

1. Identify which option (Servair, Premier, or Base) yields the highest

utility for each buyer.

2. Weight each buyer’s choice using their assigned purchase weight.

3. Compute the total weighted market share for each new product.

Then, calculate expected profit using the following formula:

Profit =300 ×Market Share ×Gross Profit per Unit

Where:

 Total Market Size = 300 units

 Gross Profit Margins:

o Servair: $180,000 per unit

o Premier: $230,000 per unit

o Base: $200,000 per unit

Finally, recommend which product Durr should launch if it can enter with only
one offering.
Part 2: Dual-Product Entry with Market Segmentation

Durr’s analysts suspect that the market consists of two distinct segments
with different preferences.
They have conducted a cluster analysis on the part-worth utilities (see the
cluster assignments in the Excel file).

Repeat the above analysis separately for each segment:

1. Compute weighted market shares of Servair, Premier, and Base

within each segment.

2. Compute segment-level profits using:

Profit =300 ×Segment Size × Share in Segment ×Gross Profit per Unit

3. Targeting strategy:
Assign one product to Segment 0 and another to Segment 1 to
maximize total profitability.
Determine which product best fits each segment based on segment
preferences.

4. Overall evaluation:

o Compare the combined profit from targeting two segments

(minus an additional cost of $3 million) with the profit from
offering a single product.

o Is the two-product strategy still more profitable?

5. Segment profiling:

o Use average part-worth utilities and firmographics (from

the discriminant data sheet) to describe each segment’s key
needs and characteristics.

o Compare the relative sizes of the two segments in terms of

their total purchase weights.

o Suggest how Durr could best reach and communicate with each
target segment.

Finally, prepare a formal business case recommending whether Durr

should:

 Enter with a single product, or

 Enter with two differentiated products targeted at the two segments.

Support your recommendation with quantitative analysis and strategic
reasoning.
Go through Durr_part_worth_utilities_clustered. Excel file
 Define the competitive set – this is the set of products from which
customers in the target segment make their choices. Some of them
may be existing products and, others concepts being evaluated. We
denote this set of products as P1, P2,...PN.

 Select Choice rule

– Maximum utility rule

– Share of preference rule

– Logit choice Rule