Data Department built a renewal and retention system for an insurance business. It combines a churn predictor with a priority-optimisation scheme that tells the team which customers to call first. The scheme is built on two models: one for churn (the chance a customer leaves) and one for propensity to buy. The result was about 6% revenue gain.
What was the decision?
Before a policy comes up for renewal, someone can call the customer. The calls take time, and a team can only make so many each day. So the real decision is a ranking: who should be called first?
Calling in policy order, or by date, treats every customer the same. They are not the same. Some customers would renew without a call. Some are likely to leave, and a call might keep them. Some are open to buying more. Time spent on the wrong call is time not spent on the right one.
The question was not "who will churn?" on its own. It was "given limited calling time, which customers should we contact first to protect and grow revenue?"
What did we build?
| Component | What it answers | How it is used |
|---|---|---|
| Churn model | How likely is this customer to leave at renewal? | Flags customers at risk |
| Propensity-to-buy model | How likely is this customer to buy? | Flags customers open to an offer |
| Priority-optimisation scheme | Who should we call first? | Ranks the call list using both models |
Why two models and not one. A churn score alone says who might leave. It does not say whether a call is worth more there or with a customer who is ready to buy. Combining churn and propensity-to-buy scores in one priority scheme lets the team weigh both and order its calls accordingly.
In general terms, priority schemes of this kind turn model scores into a ranked list that respects the team's capacity, so the most valuable calls happen first. We describe the general method here; the client's data and rules stay confidential.
Results
- About 6% revenue gain.
- A ranked call list built from churn and propensity-to-buy models, used to decide who to call first.
We report only the revenue figure for this project. We do not publish other metrics for it.
What made it work?
We started from the calling decision, not the churn model. A churn score is only useful if it changes who gets called. The priority scheme is what turned scores into action.
Retention and sales were treated together. Looking at both the risk of losing a customer and the chance of selling to them gave a fuller view of where a call adds value.
Capacity is part of the problem. A team cannot call everyone. Ranking customers within that limit is an optimisation question, not only a prediction question.
This work is part of our churn prediction and customer analytics service and our work for financial services and insurance. For another finance example, see our collections forecasting case.
Frequently asked questions
What is a churn predictor in insurance?
A churn predictor is a model that estimates how likely each customer is to leave, for example by not renewing a policy. It helps a team focus retention effort where it matters most.
What is a propensity-to-buy model?
A propensity-to-buy model estimates how likely a customer is to buy a product or offer. Combined with churn scores, it helps decide whether a call should focus on keeping a customer or on selling to them.
What result did this project achieve?
The churn predictor and the priority-optimisation scheme together produced about 6% revenue gain for the insurance business. That is the only figure we report for this project.