One of my more interesting jobs was as a manager of a credit card collections analytics group. We were tasked to produce all sorts of cool analytics, from when to schedule collectors to how to segment the delinquent customers to how to manage the collections dialer. It was fun.
Being someone who hadn’t grown up in collections, I was able to ask all sorts of stupid questions — to challenge the “traditional” way of operating. And we found some cool “traditions” that should be re-evaluated. One of the more nuanced piece of analyses we performed was to determine the priority of contacting different segments of the population.
In general, the industry had a rule of thumb when it came to prioritizing contacts: Always attempt to contact those most likely to pay. The thought went that, given that we had a limited number of resources (collection agents), we should put them on the highest odds of success.
The graphs below illustrate the point. It represents the probabilities of getting a payment from two different segments of the delinquent population. In this representative example, Segment A has the highest probability of payment, and hence should be prioritized in the dialer, right?
Well, according to our “tradition,” we would prioritize Segment A and leave Segment C as a lower priority in order to maximize the number of payments.
But here is the rub: When determining who to call, we want to judge the effectiveness of the interaction between the customer segment and the collections agent. In our made-up example, as I turn up the calling intensity, we measure the effectiveness of this interaction with the slope of this curve. As I call each of the segments more times or fewer, I can measure the effect—in payments—of the new resource allocation (more calls).
By looking at the problem this way we get a different answer. Segment C has a higher propensity of being influenced into making a payment. Because Segment A is more likely to pay without a call, it makes more sense (and makes the company more money) to call those customers who are likely to respond with payments to our contacts — not those most likely to pay.
Back in my old days, we were able to determine many of these types of segments in our customer file — the most interesting being a segment of customers who were hard to get on the phone. What we found is that these customers had never heard our collections pitch (because they were hard to get ahold of) and hence didn’t realize that our collectors were there to help. When we actually spoke to them, they reacted surprisingly well. Our data analyses showed them to respond very similarly to the Segment C group in our example.
Dan Mahon and I recently presented these sorts of ideas in a webinar “Collections Analytics: Stories about Data, Segmentation, Treatment and Scheduling,” which is now available on demand. Feel free to drop by and have a listen!