Client’s Problem: “Our sales transactions are pretty rich with data, and we don’t have any trouble reporting on that data or using canned solutions from our CRM provider to determine customer lifetime value (CLV) or predict churn. Where we’re struggling is in determining how to engage with a customer next in order to maximize our market share, not to maximize their spend necessarily, just their number of purchases. A lot of the products we sell you can get anywhere, and we want our service model to shine by getting people regularly into our sales funnel.”

This was a “fascinating” problem, which concerned us, it didn’t seem very valuable to the Client at first glance to have a consultancy dig on this problem. Market share is difficult to ascertain if the Client is not in an industry with numerous public companies that disclose data where one can reverse-engineer the market share per company. That was the case here. However, what ultimately made a solution tractable and valuable was that purchases were geospatially related.

A common approach to customer behavior modeling is to consider the purchase network or behavior tree. Customers in the Client’s industry had a set of physical limitations that constrained their purchases, adding to the behavior tree a unique set of conditions directly related to market share. As a result, we were able to build a probabilistic model for how many purchases a customer should make in a period. Combined with the Client’s existing churn and CLV models, the behavior model was able to discern missing purchases from non-existent purchases.

After segmenting and clustering the Clients’ customers, we probed clusters for actions that resulted in long-term growth in the number of transactions. We subsequently built an intelligent agent using reinforcement learning to execute the next best actions following a purchase with a high probability of increasing future transactions. The Client’s sales team used this agent to reduce the amount of unnecessary “discounting” (e.g., coupons, bundles, sales) to attract continued purchasing.