Prior to my current tour at TRC I was a partner at a small data mining boutique that had a simple objective: support the sales and marketing goals of our clients by helping them stem attrition. While the goal may have been simple, how we set out to do this was not. See, we took upwards of 30 months of time-series data on all of their customers and applied some mind-numbing statistical techniques that identified patterns in the data that preceded an eventual behavior. In most cases that behavior was a customer terminating their relationship with our client. Once our system identified the patterns that preceded attrition, we would then continue to feed it customer data each month and it would dutifully output a list of customers that were likely candidates to attrite. In addition, for each customer on this golden list we provided the prediction "trigger", or the thing that they did that was responsible for the system flagging them as a high-risk customer.
We were actually very good at building these predictive systems. Time and again we were able to achieve a prediction lift several times better than other methods our clients were using to predict attrition. So, I guess that means we were a huge success. Well, not exactly.
The fact is that a prediction by itself is worthless. Even the best marketing and sales departments need some guidance as to what to say or do to leverage the prediction's value. We thought we were providing the guidance via the behavioral triggers, i.e. the things in the data that caused a customer to get flagged as high risk. But the problem was that those triggers were backward looking. Meaning, the customer already started taking the steps of ending the relationship (which is why they were being flagged). In a very real sense, the train had already left the station.
So does this mean the attrition modeling is a bad idea? Not at all. There is great value in analyzing customer data. But in order to successfully pursue retention marketing - or any marketing for that matter - it is still necessary to take the "novel" step of actually talking to your customers. By doing so you can begin to understand what they are thinking, what they want, and what they are going to do, which can form the backbone of successful messaging. While there are certainly several ways to have these conversations, market research provides a structured, effective approach that can yield impactful customer-centric insights that go beyond other types of data. Business success (or failure) really does come down to customers and relationships, after all, not data and statistical equations.