- H2O Driverless AI can build churn and lapse models from your own policy, billing, claims and contact history, with a reason code on every score.
- A second model can predict which retention offer each at-risk policyholder is likely to accept, so incentives go where they change the outcome.
- Scores feed the CRM, agent desktop and campaign tools you already run; none of them is replaced.
- Nationwide reported more than 500,000 models instantiated and 25 billion predictions scored over a decade, across use cases that included customer retention.
Your best customers leave quietly
Your most valuable households often hold several policies, so one lapse can take a home, an auto and an umbrella policy with it. Many of them never call before they go; they simply do not renew.
Rules built on known behaviors, such as a missed payment or a recent rate increase, catch some of the leavers. They miss many others, and they send expensive incentives to policyholders who were going to stay anyway.
How AI customer retention works
H2O Driverless AI builds explainable retention models on your own data and sends the scores to the systems your retention team already works in.
Step 1: join the history
Combine policy, billing, claims, service and agent-contact records for each household, with who renewed and who left.
Step 2: predict who is at risk
H2O Driverless AI trains classification models on renewal outcomes, and automatic feature engineering finds signals such as rate change against tenure, recent claim experience and contact frequency.
Step 3: explain each score
Reason codes show which factors drove each policyholder's risk, so an agent knows what the conversation is about.
Step 4: pick the offer
A second model ranks the retention offers each at-risk household is most likely to accept, and clustering groups policyholders who leave for similar reasons.
Working inside the systems you already run
Each model ships as a scoring pipeline that runs inside your renewal process. Churn scores, reason codes and offer rankings feed the CRM, agent desktop and campaign tools your team already uses; none of them is replaced. Policyholder data stays in the environment you choose.
Results
In a published 2019 result, Nationwide reported more than 500,000 models instantiated and 25 billion predictions scored over a decade with H2O, across use cases that included customer retention.