Using a complex model for measuring success in interactions with customers

By on March 4th, 2015 in Machine Learning

The TrueAccord product can be compared to an automated marketing and sales campaign, focused on identifying payment intent and acting on it. The system classifies customers (we use “customer” to refer to those in debt) by their most probable reason for non-payment, and the “voice” we think is going to drive that to action. Then, it has to decide, out of the hundreds of content items we have, which goes out to which customer, through what channel, and when. Of course, we’d like to learn from history and send the combination most likely to succeed. This raises the important question: what’s “success” in our context?

If you run an email campaign, success is conversion – list sign-up, purchase, etc. Debt collection, however, is a much more complicated psychological process – depending on the reason for default, customers need to go through a process to accept TrueAccord’s authority, that a debt exists, and that they should pay it. This requires a compounded concept of success: not every communication converts to a payment, but many will move the customer just another step in the process.

This is what we call the “Greatest Action” model. At best, the customer presents the right frame of mind and is ready to pay immediately (our shortest communication-to-collection time is 15 minutes). Most customers, however, will go through at least a few steps before committing, similar to Kubler-Ross’s Stages of Grief:

  • Early engagement – reading an email or visiting a web page
  • Establishing that they have to deal with the debt and anger about being submitted to collections
  • Negotiation on whether, when and how to settle the debt
  • And, finally, conversion

Our model for communication success takes progression between these stages as success, to varying degrees. This allows a more nuanced view of what success is, and which interactions serve us better at every stage of the “buying process”. Email and digital communication in general lend themselves more readily for the type of immediacy and data feedback that enables this kind of interaction. As a result, the collection process resembles a customer care and retention process rather than brute force recovery, ending with better recovery and retention, as well as fewer legal issues in the long term.