Module 2

Principles and Techniques

Predictive modeling
• Predictive modeling is a mathematical process used to predict future events or
outcomes by analyzing patterns in a given set of input data.

• Predictive modeling is the process by which a model is created or chosen to try

to best predict the probability of an outcome.

• Examples of predictive modeling include the likelihood of spam or the

probability someone will click a link or buy a product.
Types of predictive models

Propensity models (predictions)

Clustering models (segments)

Collaborative filtering (recommendations)

Propensity Model
• A propensity model is a special predictive model that always
predicts Probability of a specific action. “Propensity” literally
means tendency.
• What it predicts: Probability that a user will: Buy, Click ,Churn,
Respond to an offer

• Real-world analogy
• Predictive model: Tomorrow’s temperature = 32°C
• Propensity model:Chance of rain tomorrow = 70%
Propensity Model
• Propensity Modeling is a statistical technique used to predict the chance/probability of
certain events happening in the future.
• Propensity modeling is an approach that attempts to predict the likelihood that visitors, leads,
and customers will perform certain actions.
• Propensity models make predictions about a customer’s future behavior. With propensity
models you can anticipate a customers’ future behavior.
• However, keep in mind that even propensity models are abstractions and do not necessarily
predict absolute true behavior. So they estimate likelihood, not guaranteed behavior. A
customer with 90% propensity may still not buy. A customer with 10% may surprise you.
• For example, a propensity model can help a marketing team predict, through data science or
machine learning, the likelihood that a lead will convert to a customer. Or that a customer
will churn. Or even that an email recipient will unsubscribe
Why you need Product Propensity models
1. To collect valuable insights
Propensity models analyze customer data (clicks, views, purchases, time
spent, etc.) to uncover patterns like:
• Which products people prefer
• What type of customers buy what
• Example: You may discover that users who view Product A twice are
very likely to buy Product B.
Why you need Product Propensity models
2. For tailored / strong product recommendations

Instead of showing the same products to everyone, the model

recommends items based on individual probability.

Example:

User X → 80% likely to buy Headphones

User Y → 70% likely to buy Smartwatch

Why you need Product Propensity models
3. To increase conversions
• Conversion = turning a visitor or lead into a buyer.
Propensity models help by:
• Targeting high-probability customers
• Sending offers only to people likely to respond
• Example : Instead of marketing to 100 people, you focus on the top 20
most likely to buy.
 Why you need Product Propensity models
4. To optimize resources
Marketing budgets, ads, emails, and sales time are limited.

• Propensity modeling helps decide:

• Who deserves sales calls
• Who should get discounts
• Where ads should be spent

• � You avoid wasting money on low-probability users.

• This is smart resource allocation.
 Why you need Product Propensity models
5. For churn rate analysis

Churn = customers leaving.

Propensity models predict:

• Who is likely to stop using your product

• Who may unsubscribe or cancel

Then companies can:

Offer retention discounts

Send re-engagement emails
Before Propensity Modeling (top row)
You see a mixed group of people.
This represents:
� All visitors / leads / customers treated the same.
At this stage:

•Marketing messages go to everyone

•Sales teams don’t know who is interested
•Time and money are wasted on low-interest people
After Propensity Modeling (bottom rows)
Now the same people are sorted into 3 groups based on predicted probability of responding:
� Low : People unlikely to buy / click / respond.
•Don’t spend much budget here
•Maybe nurture slowly with generic content
� Medium : People with moderate interest.
•Send targeted emails
•Provide offers or more information
� High: People very likely to convert.
•Sales team focuses here
•Personalized offers
•Priority follow-ups
 Propensity Models
Model 1: Predicted customer lifetime value (CLV)

• CLV (Customer Lifetime Value) is a prediction of all the value a business will derive
from their entire relationship with a customer.

• The Pareto Principle states that, for many events, roughly 80% of the effects come
from 20% of the causes.

• When applied to e-commerce, this means that 80% of your revenue can be attributed
to 20% of your customers.

• While the exact percentages may not be 80/20, it is still the case that some
customers are worth a whole lot more than others, and identifying your “All-Star”
customers can be extremely valuable to your business. Algorithms can predict how
much a customer will spend with you long before customers themselves realizes this.
Predicted customer lifetime value (CLV)

• Customer lifetime value helps you understand and gauge current customer
loyalty. If customers continue to purchase from you time and time again, that’s
usually a good sign that you’re doing the right things in your business.

• Customer lifetime value is a gauge of the profit associated with a particular

customer relationship, which should guide how much you are willing to invest
to maintain that relationship. That is, if you estimate one customer’s CLV to
be $500, you wouldn’t spend more than that to try and keep the relationship.

The simplest way to calculate CLV is:

• CLV = average value of a purchase x number of times the customer will buy
each year x average length of the customer relationship (in years)
 Model 2: Predicted share of wallet
It indicates what percentage of a consumer's total expenditures on a specified
product/service goes to buying Company X's brand. Share of wallet is a practical
measure of how well a company's product or service is competing in the
marketplace.

Share of wallet (SOW) is a marketing metric used to calculate the percentage of a

customer's spending for a type of product or service that goes to a particular
company.

For example, if a customer spends $60 a month at fast food restaurants, and $30 of
that amount is spent at McDonald's, McDonald's has a 50% SOW for that customer.
The term is sometimes expressed as wallet share.
Model 2: Predicted share of wallet
Case 1

Total grocery spend = $110

They spend $100 with you:

Share of Wallet = 100 / 110 ≈ 91%

� Very loyal customer.

Model 2: Predicted share of wallet
Case 2

Total grocery spend = $1,000

They spend $100 with you:

Share of Wallet = 100 / 1000 = 10%

� Huge growth opportunity — they’re mostly buying elsewhere.

Why companies care about this : Because increasing share of wallet is

often cheaper than finding new customers.
Model 3: Propensity to engage
• A propensity to engage model predicts the likelihood that a
person will engage in some activity, like unethical behavior or
post purchases.

• For example, a propensity to engage model can predict how

likely it is that a customer will click on your email links.

• Armed with this information you can decide not to send an

email to a certain “low likelihood to click” segment
Model 3: Propensity to engage
Situation
You want to send marketing emails.
Past data shows:
•Who opened emails
•Who clicked links
•Time of activity
•Purchase history
The model outputs:
Customer A → 0.82 (82% likely to click)
Customer B → 0.18 (18% likely to click)
Customer C → 0.55 (55% likely to click)

Action
•Send email to A and C

•Skip B (low engagement)

Model 3: Propensity to engage
Situation
After buying a product, some customers:
•Leave reviews
•Buy accessories
•Register warranty
The model predicts:
Customer X → 0.75 likely to buy accessories
Customer Y → 0.20 likely to buy accessories

Action
•Show accessory offers to X

•Don’t bother Y
� You upsell only where it works.
 Model 4 : Propensity to Unsubscribe
• A propensity to unsubscribe model predicts how likely it is that a customer
will unsubscribe from your email list at any given point in time.

• Armed with this information you can optimize email frequency. For “high
likelihood to unsubscribe” segments you should decrease send frequency,
whereas for “low likelihood to unsubscribe” segments you can increase
email send frequency.

• You could also decide to use different channels (like SMS, direct mail or
LinkedIn) to reach out to “high likelihood to unsubscribe” customers
Model 5 : Propensity to buy

The propensity to buy model tells you which customers

are ready to make their purchase, so you can find who to
target. Moreover, once you know who is ready and who
is not helps you provide the right aggression in your offer.
Those that are likely to buy won’t need high discounts while
customers who are not likely to buy may need a more aggressive
offer, thereby bringing you incremental revenue.

E-commerces experience very regular website traffic and every

single visitor is a potential customer. A Propensity to Buy Model
predicts which of these visitors (potential customers) are going to
purchase something and which are not.
• What Makes it Important?
E-commerce websites attract a huge amount of traffic. Ever wonder how
many actually end up buying something?

On average 2% of the visitors actually buy something. These visitors have

certain behaviors that distinguish the ones that buy something and the ones
that don’t.

These visitors can be modeled using their website visit behavior to predict
their likelihood to purchase something. This has two hidden benefits:

• The company can differentiate visitors that are highly likely to buy
something along with their behaviors.
• They can target these highly likely visitors to take immediate marketing
actions like; campaigns, coupons promotions, etc
 Model 6 : Propensity to churn
• Propensity to churn model estimates the likelihood of a customer to leave in the next
period of time.

• It uses the data about the customer, such as their service level, tenure, payment history,
as well as demographics to predict the probability of discontinuing the relationship

• It’s a binary classifier that divides clients into two groups (classes) — those who leave
and those who don’t.

• In addition to assigning them to one of the two groups, it will typically give us the
probability with which the client belongs to that group
It is important to note that this is the probability of
belonging to the group of clients who leave.

• What is it useful for?

By knowing which clients are at the highest risk of leaving,
we can better target our rescue efforts.

For example, we can reach out to these clients with a

marketing campaign, reminding them that they haven’t
purchased from us in a while, or even offering them a
benefit.
The propensity to churn model tells you which active customers
are at risk, so you know which high value, at risk customers to put
on your watch list and reach out.
Armed with this information, you may be able to save those
customers with preemptive marketing programs designed to retain
them.

Often propensity models can be combined to make campaign

decisions. For example, you may want to do an aggressive
customer win back campaign for customers who have both a high
likelihood to unsubscribe and a high predicted lifetime value
Cluster Models

• Clustering is the predictive analytics term for customer segmentation.

Clustering, like classification, is used to segment the data.

• Unlike classification, clustering models segment data into groups that were
not previously defined

• Cluster analysis itself is not one specific algorithm, but the general task to
be solved.

• It can be achieved by various algorithms that differ significantly in their

notion of what constitutes a cluster and how to efficiently find them.
Clustering models focus on identifying groups of similar
records and labelling the records according to the group to
which they belong.

While clustering is useful for data analysis and as a pre-

processing step for a number of learning tasks, we are
interested in the specific pre-processing task of using
clustering to gain more information about the data to
improve prediction accuracy.
Model 1: Behavioral clustering

Behavioral clustering informs you how people behave while purchasing.

Do they use the web site or the call center? Are they discount addicts?
How frequently do they buy? How much do they spend? How much time
will go buy before they purchase again?
Example: Online Fashion Store
Suppose an online fashion brand analyzes its customer data and creates behavioral clusters.

1️ Discount Addicts
Always purchase during sales
Use coupon codes frequently
Abandon cart if no discount is available

2. Premium Loyal Customers

Buy new arrivals at full price
Rarely use discounts
High average spending
Purchase every 1–2 months

3. Frequent Small Buyers

Purchase every 2–3 weeks
Spend small amounts each time
Mostly buy basic items
Example: Online Fashion Store

Key Behavioral Variables Used in Clustering

Purchase frequency

Average order value

Time between purchases

Discount usage
• This algorithm helps set the right tone while contacting the customer.
For instance, customers that buy frequently but with low sized orders
might react well to offers like ‘Earn double rewards points when you
spend $100 or more
• Behavioral clustering can also informs us on other behaviors,
such as crime and is used in performing crime analysis
 Model 2: Product based clustering
(also called category based clustering)
• Product based clustering algorithms discover what different groupings
of products people buy from.
• See the example below of a category (or product) based segment or
cluster. You can see people in one customer segment ONLY buy Pinot
Noir, whereas those in another customer segment buy different types of
Varietal products, such as Champagne, Chardonnay, Pinot Grigio and
Prosecco - but never Cabernet Sauvignon, Malbec or Espumante.

• This is useful information when deciding which product offers or email

content to send to each of these customer segments.
• The following use cases describe how product-based clusters help you
produce relevant and personalized offers, thereby increasing customer
engagement and reducing marketing costs

• More targeted product or brand campaigns When creating a product-

focused campaign, quickly identify and target the customers or clusters that
have previously purchased these products. Such campaigns help to increase
affinity toward the products that customers already likeWhen launching a new
product, identify a current similar product, locate the customer cluster aligned
with the existing product, and target customers in those clusters in the launch
campaign.

• This ensures a high response rate because you target customers with similar
product affinities.
• Create surgical discounting or pricing tests:

For example, you might test giving a lower discount for product X
(or similar products) to customers that belong to cluster (s) that are
aligned to product X. If a customer already likes product X, the
customer does not need the incentive of a high discount to buy
product X (or products similar to it).
Model 3 : Brand Based Clustering
• It focusses on the brands of the products that the customer buys.
• Marketers can use this information to project what other brands those
customers are likely to buy.
• Customers are then ordered according to Nike, Adidas, Under Armour,
etc. Now you know what specific brands to pitch to certain customers.
When a brand releases new products - you know who is likely to be
interested.
 Collaborative filtering

• Collaborative filtering is a technique used in recommender

systems to predict a user's preferences or interests based on
the past behavior or preferences of similar users.

• It is based on the assumption that people who have similar

interests in the past are likely to have similar interests in the
future
 Cont..

• In collaborative filtering, the system first collects information

about the preferences and behaviors of a large number of
users.
• Then, it identifies users who have similar preferences and
behaviors, and makes recommendations to a user based on the
interests and behaviors of those similar users.
 Cont..
• There are two main types of collaborative filtering: user-
based and item-based.
• In user-based collaborative filtering, the system
recommends items that similar users have liked in the
past.

• In item-based collaborative filtering, the system

recommends items that are similar to the ones a user
has liked in the past.
Cont…

• Collaborative filtering has become a popular technique for

recommendation systems used in e-commerce, social media, and other
industries.