Finding the most likely repurchasing customers next month: For targeted promotions

Project Details


Fall 2019


Rona Lu-Lai, Vivian Lu, Joanne Hsueh, David Hung





Business Problem

Our business goal is based on the current marketing strategies of Me-come. Most of the current marketing strategies of Me-come were not targeted at specific groups of customers. For example, Me-come sends all customers text messages or send flyers to nearby residents, which is costly and inefficient.


Our analytical objective is to sort VIPs by their potential to repurchase next month and to identify the top 10% of VIPs who most likely to return. VIPs with at least one transaction record in the next month will be classified as returning customers(1), the others will be classified as not returning customers(0). We used VIP, Transaction and Product data from Me-come and selected 77820 VIP customers as our data.

Analytics Solution
We found that total sales points, purchase times Dec-May, pur times morning-evening(all purchase times), min/mean/max sales amount, category 1/5/7 are important factors. These factors are related to the purchase frequency and sales amounts of each VIP. It implies that VIPs who purchase more times or amounts will have a higher probability to return next month.

Best Model on test data

The best model is a classification method called Random Forest with important factors mentioned above. By using our model, Me-come can catch 1.83 times of repurchasing VIPs than randomly selecting by only sending promotions to 10% of VIPs.


We shall forward the result to our marketing team after our monthly expected results so that they can establish marketing strategies. The marketing team will provide the sales team with tactics for the implementation and will deliver promotions to potential customers.

Production issues
Since we use historical data, we will not run our solution in real-time. We need to collect and update new purchase information on a monthly basis.


Application Area: