Use case · Insurance
Cancellation early warning for insurers: spot at-risk policies before the notice arrives
Cancellation early warning is a prediction of which insurance policies are likely to be cancelled in the coming weeks. UNID AI Predict delivers a weekly ranking per adviser, weighted by premium at risk, with the reasons from the data and the next best products.
Last updated:
The problem
By the time the cancellation notice arrives it is usually too late. At the same time, sales cannot call every customer. What is needed is a short list with reasons: whom first, and with which offer.
What you get
- A weekly worklist per adviser, sorted by risk × annual premium
- For each policy, the facts that drive the score
- The top three next products per customer (next best product)
- A backtest you recalculate yourself
Animated sample view: a worklist shows the policies with the highest cancellation risk in the next 90 days, sorted by premium at risk. Clicking a policy shows the reasons from the data and the next best products. A backtest shows how the insurer checks the prediction themselves.
What data is needed
Seven tables are enough to start. Names and addresses are not needed; customers can be pseudonymised.
| Table | Content (examples) |
|---|---|
| Customers | Customer ID, age band, region, customer since |
| Policies | Policy ID, line, start, premium, changes |
| Cancellations | Policy ID, date, reason (if recorded) |
| Premium invoices | Amount, due date, payment, reminders |
| Claims | Notification, status, settlement, decline |
| Advisers | Adviser ID, responsibility, changes |
| Products | Product catalogue and lines |
How the prediction is checked
Hidden cut-off date
You provide data up to a cut-off date and keep the cancellations after it.
Baseline
Your existing model or a simple rule (e.g. “premium increase above X%”).
Metrics
Hit rate and lift in the top 1, 5 and 10% of the list.
Value calculation
With your own save rate and contact costs, not our assumptions.
We do not promise an improvement in percent up front. You measure it in the backtest on your data.
Data protection and governance
- No health data needed
- Marketing consent respected
- Never used for pricing or underwriting decisions
- A person decides on every contact
- Log, model card, fairness report by age band and region
- Works council involvement planned
Deployment
Predict runs as a container on the insurer’s infrastructure, with a local language model and no outside connection if required. Alternatively on a dedicated server with a processing agreement.
What it does not do
- The prediction says who is at risk, not how to keep them. That remains the job of sales.
- Without cancellation history covering at least one period as long as the prediction window, no statement is possible.
- Against a well-tuned existing model the margin can be small.
FAQ
Frequently asked questions
What is cancellation early warning?
What data does churn prediction need?
Does customer data have to leave the building?
How much better is this than our existing model?
Is this a high-risk system under the EU AI Act?
What is next best product?
Which lines of business are suitable?
See it working on your own material.
In a 30-minute call we show the products live and work out which use case makes sense for you first. No obligation, in German or English, in Vienna or online.
UNID AI · Leonard-Bernstein-Straße 10 · 1220 Vienna · +43 660 476 30 89