Use AI To Predict Who To Target Offers At
How to avoid Coupon Hunters by predicting the long-term impact of offers on each shopper
Most Important Features In Coupon Uplift Prediction
“Coupon Customers” often don’t pan out
Acquiring new customers through coupons or offer can seem very lucrative — at first. But then these customers often don’t return. They are just hunting for the next offer somewhere else.
Prediciting likelihood to purchase is not enough
Most AI systems make the mistake of just predicting who will use a coupon. That can lead to a negative feedback loop: You will laser focus your campaigns on the most “coupon hungry” audience. And those then also end up having the lowest Customer Lifetime Value (CLTV). That’s not good.
AI can predict who is actually worth targeting
The challenge is to predict which customer will return after using the coupon. You should aim for those that return and become a long term customer. With the right setup Machine Learning can also help you to find out what defines those special shoppers. Once you trained the system the right way, you can focus only those segments.
Data you need: Purchase histories
You need the purchase histories for a large number of customers who were targeted for a campaign. Specifically, their purchase history before and after the coupon campaign. You also need to have recorded which offer each customer received when.
How to prepare the data?
Features that are good at predicting good coupon campaign targets:
- The number of times the customer purchased a discounted product before
- The total sum of purchases of that customer
- Whether the customer has ever bought something before
- The number of times a customer has bought the brand on offer before (and on offer)
- Whether the customer has bought that brand before, and how often
- How big the offer discount is
- The number of products that the offer can be used on
Most important criteria: Has the customer purchased that brand before?
GLMs perform well
An algorithm that perform well on this problem: Lasso and Elastic-Net Regularized Generalized Linear Models (GLMs) - See glmnet
Predicting repeat purchases is somewhat a subproblem of Customer Lifetime modelling:
- Modeling Customer Lifetime Value
- Measuring Customer Life Time Value: Models and Analysis
- [A modified Pareto/NBD approach for predicting customer lifetime value(http://www.essec.edu/faculty/showDeclFileRes.do?declId=8555&key=Publication-Content)
- Improved Pareto/NBD Model and Its Applications in Customer Segmentation based on Personal Information Combination
Need help with a machine learning challenge?Schedule a Call
OR send me a description to: firstname.lastname@example.org