Our marketing team can see who has left (descriptive), but we are failing to understand why and what we should do to stop it (prescriptive). What are the steps to build a prescriptive engine that suggests specific retention offers based on a customer's behavior profile? We are currently using SQL and Tableau but feel like we need a more advanced machine learning approach.
3 answers
Moving to prescriptive analytics requires a solid foundation in predictive modeling first. You need to build a model that assigns a "churn probability score" to every customer. Once you have that, you apply an optimization layer. For example, if a high-value customer has a 70% chance of leaving, the prescriptive engine might suggest a 20% discount. If they are low-value, it might suggest just an email check-in. This involves using Decision Trees or Random Forest algorithms to determine which "treatment" has the highest likelihood of a positive outcome based on past success.
Have you considered the ethical implications of these automated offers, specifically regarding price discrimination among different customer segments?
Start by mapping out your "Customer Journey." You can't prescribe a cure if you don't know at which touchpoint the customer experience is actually breaking down.
Alice is spot on. A prescriptive model is only as good as the journey data you feed it. Without touchpoint data, the suggestions will be generic.
Thomas, that is a very valid concern for our legal team. We plan to keep the "prescriptive" part as a recommendation for our human account managers first, rather than fully automating the offers. This "human-in-the-loop" approach ensures that we aren't inadvertently penalizing loyal customers while trying to save at-risk ones. We want to use the analytics to empower our staff, not replace their judgment entirely, especially in B2B.