Mathematics: Analysis and Approaches Higher Level

Internal Assessment

Exploring the use of Bayes’ Theorem in Marketing with the

Case Study of Taget’s Pregnancy Prediction Score

 2

Table of Contents

Title Page......................................................................................................................................... 1

Table of Contents............................................................................................................................. 2

Introduction......................................................................................................................................3

Mathematical Background and Derivation...................................................................................4-9

Target Pregnancy Prediction Case Study..................................................................................10-12

Outcomes of Target’s use of Bayes’ Theorem..................................................................... 13-asasd

Conclusion................................................................................................................................ asdad

Works Cited......................................................................................................................... asdasdas

 3

Introduction:

Businesses rely heavily on data and predictive probability to infer customer behavior and

make decisions regarding marketing strategies. One of the most powerful probability tools is

Bayes’ Theorem, a method that updates probabilities of future events (or customer profiles) with

regards to the past. Companies like Amazon, Google, Netflix, and Target all use this tool of

predictive analytics to personalize recommendations, optimize ad targeting, and even predict

major life events such as pregnancies in their customers.

My topic is Bayes’ Theorem and the applications of that theory to predict future

consumer behavior. My interest in this topic came from a summer e-marketing class I took in

Hong Kong. I was fascinated by how marketers could predict future behavior and send targeted

ads. The professor used an example based upon the retailer Target’s customer profiling method

that was able to accurately predict that a customer was pregnant based solely upon what that

person purchased. I later found an article written by Kashmir Hill for Forbes about how Target

could predict a pregnancy before even the woman knew. This then combined with my interest in

behavioral economics, my college major, and was the motivation for writing this IA about

customer segmentation and Bayes’ Theorem. The real-world applications of mathematics, and

specifically probability theory fascinate me which leads me to write this investigation about how

Bayesian inference can be used to model and predict consumer behavior. In this paper, I seek to

understand how and why Target can so accurately predict customer behavior using only purchase

data.

Therefore, the research question of this investigation is: “How can Bayes’ Theorem be

used to mathematically model and predict customer profiles of purchasing behavior for targeted

segmentation in marketing?”
 4

Mathematical Background and Derivation of Bayes’ Theorem:

To start, the definition of Bayes’ Theorem is as follows:

𝑃(𝐵|𝐴)𝑃(𝐴)
𝑃(𝐴|𝐵) = 𝑃(𝐵)

This equation allows us to invert the probability of event B given that event A into the

probability of event A given that event B has already occurred. This is known as a conditional

probability, or the posterior probability of A given B.

P(A) is the probability of event A happening, or the marginal probability. It is called the

marginal probability because it is the probability of event A without any knowledge of event B.

P(B) is the probability of event B and is also the marginal probability of event B without

knowledge of event A. This is the likelihood of an event regardless of the outcome of another

variable.

P(B|A) is a conditional probability, like P(A|B), because it requires event A to have

occurred. P(B|A) is the probability of event B occurring given that A is true. Note, that because

conditional probabilities hinge on an event occurring, the probability of that prior event cannot

be zero. For instance, there can be no P(A|B) if event B has not occurred.

By definition, the probability of event A occurring given that event B has occurred is the ratio of

the joint probability of both events A and B occuring to the marginal probability of event B:

𝑃(𝐴∩𝐵)
𝑃(𝐴|𝐵) = 𝑃(𝐵)
 5

Similarly, the probability of event B occurring given that event A has occurred is the ratio of the

joint probability of both events A and B occuring to the marginal probability of event A:

𝑃(𝐴∩𝐵)
𝑃(𝐵|𝐴) = 𝑃(𝐴)

These conditional probabilities rely on joint probabilities (ex. P(A∩B) and P(B∩A),

where we know P(A∩B)=P(B∩A) ) which are the probabilities of both events A and B

occurring.

These joint probabilities are written as 𝑃(𝐴 ∩ 𝐵) using the equations:

𝑃(𝐴 ∩ 𝐵) = 𝑃(𝐴|𝐵)𝑃(𝐵)

𝑃(𝐴 ∩ 𝐵) = 𝑃(𝐵|𝐴)𝑃(𝐴)

Then these equations are set equal to each other:

𝑃(𝐴|𝐵)𝑃(𝐵) = 𝑃(𝐵|𝐴)𝑃(𝐴)

Then P(A|B) is isolated:

𝑃(𝐵|𝐴)𝑃(𝐴)
𝑃(𝐴|𝐵) = 𝑃(𝐵)
 6

Note that this is the definition of Bayes’ Theorem.

However, Bayes’ Theorem can be generalized using the Law of Total Probability:

𝑛
𝑃(𝐵) = ∑ 𝑃(𝐵|𝐴𝑖)𝑃(𝐴𝑖)
𝑖=1

The Law of Total Probability is the total probability of event B, given that event B can happen

due to several probable causes which are represented using 𝐴𝑖.

​ To explain this law, imagine there are 2 bags of red and blue marbles:

●​ The probability of drawing a blue marble is P(B)

●​ The probability of drawing from bag 1 is P(𝐴1) = 0.6

●​ The probability of drawing from bag 2 is P(𝐴2) = 0.4

●​ Bag 1 and Bag 2 have different numbers of blue and red marbles, thus have

different probabilities for drawing a blue marble.

●​ The probability of drawing a blue marble in bag 1 is written as P(𝐵|𝐴1) = 0.3, or

the probability of drawing a blue marble, given that bag 1 has been chosen is

30%.

●​ Likewise, the probability of bag 2 is written as P(𝐵|𝐴2) = 0.5

To find the overall probability of drawing a blue marble (P(B)), regardless of which bag picked

one would use the expression:

𝑃(𝐵) = 𝑃(𝐵|𝐴1)𝑃(𝐴1) + 𝑃(𝐵|𝐴2)𝑃(𝐴2)

 7

𝑃(𝐵) = (0. 3)(0. 6) + (0. 5)(0. 4)

𝑃(𝐵) = 0. 18 + 0. 2 = 0. 38
Essentially, this expression is just stating that the total probability of drawing a blue

marble is equal to the probability of choosing a blue marble in bag 1 multiplied by the

probability of drawing from bag 1 added to the probability of drawing a blue marble in bag 2

multiplied by the probability of choosing bag 2.

To condense that statement, it is the total probability of an event happening (drawing a

blue marble), given that another prior event (whether bag 1 or 2 is chosen) influences the

probability of that event (drawing a blue marble in that respective bag).

Using the law of total probability, Bayes’ Theorem can be rewritten as:

𝑃(𝐵|𝐴𝑘)𝑃(𝐴𝑘) 𝑃(𝐴𝑘∩𝐵)
𝑃(𝐴𝑘|𝐵) = 𝑛 = 𝑛
∑ 𝑃(𝐵|𝐴𝑖)𝑃(𝐴𝑖) ∑ 𝑃(𝐵|𝐴𝑖)𝑃(𝐴𝑖)
𝑖=1 𝑖=1

𝑃(𝐴𝑘∩𝐵)
= 𝑛
∑ 𝑃(𝐴𝑖∩𝐵)
𝑖=1

𝐴𝑖 is the representation of all possible causes that could lead to event B. 𝐴𝑘 is the specific

event that has been singled out. To exemplify this generalized rule, imagine now drawing a blue
 8

marble from one of 3 bags, each bag has a different probability of being chosen, and the ratio of

red to blue marbles are different in each bag.

This concept is most easily demonstrated through the use of a tree diagram:

This is a general explanation of Bayes’ Theorem and its generalization. However, this

investigation uses the example of marketing. Very generally, companies leverage prior customer

data (event B in my initial equation) to then create a Bayesian Inference to anticipate the needs

of customers and personalize marketing strategies (event A).

 9

A Note about Citations: All examples (numbers, diagrams, equations) for this section

were fully independently generated. Though no quotes have been used, I did take inspiration for

the explanation and diagrams from the works of Donald Berry, Greg Allenby, and Peter Rossi

(all of which are cited below).

 10

Target Pregnancy Prediction Case Study:

​ Humans are creatures of habit, and they tend to follow these habits very predictably.

However, whenever a major life event occurs, there are certain windows of time for those habits

to change. One particularly notable example is when someone is having a baby. During

pregnancy, women particularly experience a significant period of change, physically and

psychologically which reflects in their buying habits and increases their flexibility to change

their spending habits (Duhigg The Power). For example, pregnant women are more likely to

become allergic or sensitive to product additives, so are more likely to buy fragrant free or

hypoallergenic versions of a product. Changes in purchasing patterns like this are picked up as

potential indicators of pregnancy. Depending on the items being purchased, the retailer may even

be able to speculate what trimester of pregnancy the person is in.

Because pregnancy is an event that makes routines pliable, it is imperative for companies

to capitalize on this newfound flexibility. One way the retailer Target found to get ahead of its

competitors was to use Bayes’ Theorem to predict pregnancy. It sought to identify and direct

targeted advertisements to pregnant people promptly, to get ahead of competitors and offer better

deals to the consumer throughout pregnancy. To do this, Target analyzed the historical data it had

collected from its consumers which was simple because Target assigns every customer a specific

Guest ID number. This number is tied to a customer’s name, credit card, email address, etc.

which stores a history of everything that person has bought.

Target then analyzed what markers were common in people who were later confirmed to

be pregnant. These markers were demographical like age and gender, but also based on purchase

history. Through an analysis of this data, they found that pregnant women often bought:
 11

unscented lotions, vitamin supplements, cotton balls, hand sanitizers, and other similar products.

These purchases became predictive signals which could indicate a pregnancy.

​ Given these common purchases, Target’s statisticians built a predictive model to estimate

the likelihood that a customer was pregnant. This model assigned a “pregnancy prediction score”

based on the purchase history of a given customer which employs Bayes’ Theorem to make a

Bayesian Inference which updated the probability of a customer.

A simple version of this calculation might look something like this:

𝑃(𝐵|𝐴)𝑃(𝐴)
𝑃(𝐴|𝐵) = 𝑃(𝐵)

●​ Probability A: The likelihood that a customer is pregnant

●​ Probability B: The probability a customer buys one of the identified pregnancy

related products

●​ Probability of B|A is the likelihood of buying a given product given the customer

is pregnant

According to Bayes’ Theorem, a store can predict the probability that a customer is

pregnant, given the purchase of a pregnancy-related item. This can be calculated because of

Bayes’ Theorem and the fact that prior data gathered from the large customer base can be used

and future information can be inserted into this equation. For example, the probability A of a

customer being pregnant can change based on age, gender, and location of the store.

Furthermore, the probability of a customer purchasing a pregnancy related product can be

 12

updated by adding or subtracting items from the list of pregnancy related products which will

influence the probability that a pregnant person will purchase a pregnancy linked item.

​ For each additional purchase from the pregnancy-related list, the posterior probability

was updated, increasing the confidence in the prediction. This confidence in this prediction can

be refined to even estimate the gestational stage of a pregnancy and due date.

​ An example of this is that if a customer purchases prenatal vitamins for the first time

followed by maternity clothes months later, the model adjusts the estimated due date to allow

Target to time their advertisements. Ex. sending diaper coupons around the expected due date

rather than risk being too early or too late (both of which would be a waste of resources to the

company and unhelpful to the consumer).

​
 13

Outcomes of Target’s Pregnancy Prediction:

​ There were many results of Target’s implementation of Bayes’ Theorem to predict

pregnancy and target advertisements. The first outcome of this Pregnancy Prediction algorithm

was its accuracy. Andrew Pole, a head statistician for Target, was quoted saying that the

pregnancy prediction score “... was remarkably accurate, estimating a woman’s due date with a

staggering 90% precision” (Akande). Through this case study, Target demonstrated how truly

powerful, and accurate Bayesian inferences could be.

Given this high accuracy, Target’s Guest Analytics division was nothing short of staggering with

an estimated additional $600 million per year to Target’s revenue.

​ However, there were serious ethical concerns about customer privacy and transparency.

Strategies to avoid creepiness - not sending only baby-related coupons to make the ads feel less

target which reduced the creepiness factor while still maintaining marketing effectiveness

​ Integrating Bayesian methods has enabled companies like Target to more precisely target

and identify customer needs to personalize marketing efforts. This use of Bayes’ Theorem

increases the effectiveness of marketing campaigns and optimizes the allocation of resources.

According to Harry V. Roberts in the Journal of Marketing, Bayesian statistics provide a robust

framework for decision making under uncertainty, allowing marketers to incorporate prior

knowledge and continuously update their strategies based on new data.

​ The success around predictive analytics have prompted a broader industry reflection on

data usage with companies now being more cautious about measures to anonymize data to

minimize backlash and lessen the creep factor

 14

Results and outcomes

-​ Implementation of Predictive Analytics

-​ Customer Reactions

Ethical considerations

-​ Privacy Concerns

-​ Corporate Responsibility
 15

Conclusions:

​ This investigation has explored how Bayes’ Theorem can be derived with two or

infinitely many events.

​ Bayes’ Theorem can be applied to predict customer purchasing behavior which can be

leveraged in marketing strategies exemplified through the marketing strategies used by Target to

identify expectant mothers. Through an analysis of Bayesian probability, I have sought to

demonstrate how prior probabilities can be updated with new data to refine predictions.

​ Companies leverage this approach to optimize targeted advertising, improving both the

efficiency of their marketing efforts and customer engagement.

​ A key focus of this study is how Bayesian methods allow businesses to make highly

accurate predictions based on patterns in consumer behavior.

However, this investigation is significantly limited in scope. There are further

applications in how these predictive models can be used to retain customers through loyalty

programs and identify future customer value (or lifetime value). Furthermore, this investigation

only focuses on the sector of marketing and customer segmentation. There are further

applications of the Bayesian Inference such as in finance to assess credit risk and detect fraud, in

healthcare to diagnose diseases, and in machine learning or artificial intelligence to enhance

personalized recommendation algorithms.

In conclusion, Bayesian probability provides a powerful framework for predicting

consumer behavior with a high degree of accuracy. The learning from seemingly random events

to yield accurate predictions may be unsettling customers.

16
 17

Works Cited

Akande, Toba. “Target Predicted It Customer's Pregnancy and Made Billions.” Medium,

Medium, 2 Oct. 2023,

[Link]/target-predicted-it-customers-pregnancy-and-made-billi

ons-578131e83375. Accessed 11 Mar. 2025.

Allenby, Greg M., and Peter E. Rossi. “Teaching Bayesian Statistics to Marketing and Business

Students.” The American Statistician, vol. 62, no. 3, 2008, pp. 195–98. JSTOR,

[Link] Accessed 11 Mar. 2025.

Assmus, Gert. “Bayesian Analysis for the Evaluation of Marketing Research Expenditures: A

Reassessment.” Journal of Marketing Research, vol. 14, no. 4, 1977, pp. 562–68. JSTOR,

[Link] Accessed 11 Mar. 2025.

Berry, Donald A. Statistics: A Bayesian Perspective. Duxbury Press, 1996.

Duhigg, Charles. “How Companies Learn Your Secrets.” The New York Times, The New York

Times, 16 Feb. 2012, [Link]/2012/02/19/magazine/[Link].

Accessed 11 Mar. 2025.

Duhigg Charles. The Power of Habit. Memories of Ages Press, 2021.

Green, Paul E., and Ronald E. Frank. “Bayesian Statistics and Marketing Research.” Journal of
 18

the Royal Statistical Society. Series C (Applied Statistics), vol. 15, no. 3, 1966, pp.

173–90. JSTOR, [Link] Accessed 11 Mar. 2025.

Hill, Kashmir. “How Target Figured out a Teen Girl Was Pregnant before Her Father Did.”

Forbes, Forbes Magazine, 20 Feb. 2024,

[Link]/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pr

egnant-before-her-father-did/. Accessed 11 Mar. 2025.

Wu, Jiexing, et al. “Bayesian Inference for Assessing Effects of Email Marketing Campaigns.”

Journal of Business & Economic Statistics, vol. 36, no. 2, 2018, pp. 253–66. JSTOR,

[Link] Accessed 11 Mar. 2025.