Smart Marketing and Strategising with a Propensity Model for user conversion on their return visit | DataScience in Ecommerce and Foodtech Platforms

Abstract

Marketing is all about customer pull strategy. Companies burn a lot of cash luring customers with offers, running Ad campaigns, sending emailers, and push notifications to improve their conversion numbers. Marketing Teams in every company invest a lot of time talking about the significance of getting the correct messages to the perfect individuals at the perfect time.

Notifying or Emailing when the user is not interested may cause many users to turn off app notifications or report emails spam or worst even uninstall the app which blocks all future communications.

Marketing comes at a cost both financial and user experience. It is being wise to put in effort for only a subset of users who might be interested to purchase or have a high probability to convert.

The best way to identify?who among your audience is most likely to make a purchase, accept an offer, or sign up for a service is a?propensity model.

Let us understand the propensity model better by working on a problem statement:?Build a propensity model to determine if a user will purchase on their return visit.

I have tried to explain the concept and thought process with a sample synthetic dataset. This model can be implemented on the majority of B2C platforms: FoodTech or Ecommerce or Edtech mainly.

Goal

Objective

• Propensity model to figure out the likelihood of a person buying a product on their return visit.
• We need to identify the probability to convert for each user.

Business Outcome

• Propensity score would help segment users and run different marketing campaigns so that the conversion rate increases.
• We may share mobile notifications or emailers with exclusive offers (discounting) to high propensity users helping businesses with conversation rates.

Dynamic pricing, Coupon discounts, or exclusive offers on products shown to a high propensity on their return visit on the app may lead to better conversions.

Dopamine Effect and Feel Exclusive Strategy (Image: User journey and behavior with promotional offers)

Approach

What is a Propensity Score?

• A propensity is a natural tendency to behave in a certain way. We all have propensities — things we tend to do.
• Propensity modeling attempts to predict the likelihood that visitors, leads, and customers will perform certain actions.
• Propensity score can be defined as P( Target = 1 | Z ) where Z is the user’s behavior characteristic features on the website.

Feature Required

Data Analysis

Analyze your data understanding how many % of the positive class (user buy on a return visit) and negative class (user doesn’t buy on a return visit).

Positive Class (1) i.e. user buy on return visit: 1.53% [High Propensity Customer]

Negative Class (0) i.e. user doesn’t buy on a return visit: 98.47% [Low Propensity Customer]

Metric Selection

For our Marketing use case to improve conversion rate:

Cost of False Negative (predicting High Propensity as Low) > Cost of False Positive (predicting a Low Propensity Customer as High)

Hence our Metric should be such that: Recall is more important than Precision

A beta value of 2 will weigh more attention on recall than precision and is referred to as the F2-measure.

• F2-Measure = ((1 + 22) * Precision * Recall) / (22 * Precision + Recall)

Secondary Metric: AUC-ROC

Model Training

The propensity model is a binary classification problem, we would be using a Logistic Regression for our model. (Below Image is Feature Schema for Model)

Model Output

prob:?is?logistic regression?probability?of an event occurring, in our case event is user buying on a return visit or not.

Experimentation Table

We ran 3 different feature set experiments with logistic regression and found 2nd to be performing best on our metrics.

VISUAL of Model Evaluation (Best Model: 2nd in above Experiment Table | Positive Class Threshold: 0.0217)

The best threshold for positive class = 0.0217 means logistic regression probability ≥ the threshold is positive class (user will buy on return visit) else, negative class.

Results on Test Dataset for Propensity Model

On testing experiment model 2 with features Bounce, OS, TimeOnSite, Pageviews, and Country. We got a Recall of 91.7% and a Precision of 3.9%. A high recall relates to low False Negative cases and low precision relates to high False Positive cases.

Confusion Matrix on the Test dataset

NOTE: To build this model our objective was to maximize the conversion rate. We gave more importance to recall i.e. Cost(False Negative) > Cost(False Positive)

If marketing communication cost is high and business demands (equal Precision and Recall) then we would need to change the positive class threshold and metric such that Recall = Precision (take F1 Score as the metric).

Scope of Improving the Model

• If the user is signed in on the website or not.
• Add Traffic Source; organic or ads brought them to the website
• Use time features: some users love shopping at night (case study link )
• Use Technographic Features: Device Brand and Model to target users better. There have been multiple case studies that back that users having costly/high-end devices convert more. (case study link )
• Use Demographic features (Age, Gender, etc.) if any captured at the time of sign up of user to the platform
• Number of days gap b/w first visit and return visit to the app
• Consider experimenting with the model by adding more geographics of the user i.e. City/State. Users visiting the website from Tier 1 cities or state having higher per capita have a higher propensity to convert.

Conclusion

Now, using this propensity model marketing, dynamic pricing, product discounting, and audience targeting can be done more intelligently where chances of a user conversion (purchase) from the platform are higher. Also, it helps the marketing and user acquisition team in terms of cost as they no longer have to run campaigns/notifications/emailers on all visitors rather focus only on a subset of users whose propensity score is high.

This newsletter is now being read by more than?3k+?subscribers, please comment below your suggestions and help share this with more people, have a great weekend ??Shaurya

I hope you learned something new from this post. If you liked it, do subscribe, hit ?? or ??, and share this with others. Stay tuned for the next one!

Connect, Follow or Endorse me on?LinkedIn ?if you found this read useful. To learn more about me visit:?Here