UK Bank — Using Propensity-to-Pay to Focus Collections Effort
How analytics can improve collections strategy when the real problem is not lack of effort, but where that effort is directed.
This page continues the same transformation case in depth. It covers the situation, transformation logic, implementation considerations, practitioner lessons and a reusable framework.
← Read the transformation in under a minuteWhat the case teaches when you look beneath the headline.
1. The situation
Collections operations often appear to have a capacity problem: too many accounts, too few agents, not enough time. But increasing capacity is not the only answer. The deeper question is whether the existing capacity is being directed intelligently.
2. Analytics changes the queue, not just the report
A propensity-to-pay model becomes operationally useful only when it changes who is contacted, when, through which treatment path and with what level of effort.
3. Segmentation must connect to action
Segments are not outcomes. Each segment needs a treatment logic. High-propensity customers may require different timing or channel strategies from customers who are less likely to respond.
4. Avoid turning the model into a black box
Operations leaders need to understand enough of the logic to trust and govern it. They also need to monitor whether the model continues to discriminate meaningfully as customer behavior changes.
5. Measure both model performance and business performance
Model accuracy is not the same as collections improvement. The operational test is whether prioritization improves recovery outcomes, agent productivity or customer treatment compared with the previous strategy.
What practitioners can reuse.
Analytics must alter a decision
A model that never changes workflow is an insight artifact, not a transformation.
Segments need treatments
Prioritization becomes useful when each segment has an operational action attached.
Operational trust matters
Frontline adoption improves when people understand how model outputs should influence their work.
Monitor outcome drift
A model can remain technically functional while losing business relevance over time.
A practical way to approach a similar problem.
- Define the decision — Specify what the model is supposed to improve: order, channel, timing or treatment.
- Build the outcome label — Choose a meaningful definition of repayment / response.
- Create propensity segments — Convert raw scores into operationally useful groups.
- Design treatment paths — Define what the operation should do differently for each segment.
- Pilot against a baseline — Compare results with the existing prioritization approach.
- Monitor drift and value — Track both model performance and collections outcomes.
Use the case as a discussion guide.
- What decision will the propensity score change?
- Can agents explain how different segments should be treated?
- Are we measuring recovery outcomes or only model accuracy?
- What happens when customer behavior changes?
- Could our prioritization unintentionally create poor customer treatment?
Return to the one-minute transformation summary.
← The transformation in under a minute