Policyholder Retention Prediction: Identifying Flight Risk 90 Days Early

A personal auto carrier with a 12% annual non-renewal rate on a 500,000-policy book loses 60,000 policyholders per year. At an average premium of $1,200, that is $72 million in annual premium walking out the door. Replacing those policies through new business acquisition costs the carrier roughly $400-$600 per policy in marketing, commission, and underwriting expenses. The annual cost of attrition is between $24-36 million, and that does not count the lost underwriting profit on the departed policies.

If the carrier could identify the 60,000 flight risks 90 days before their renewal and save even 15% of them through targeted retention efforts, they would retain 9,000 policies worth $10.8 million in premium at a fraction of the cost of acquiring replacement policies.

What Predicts Non-Renewal

Retention prediction models identify the factors that correlate with non-renewal based on historical data. The predictors fall into several categories.

Pricing factors are the most obvious. Policyholders who received significant rate increases at their last renewal are more likely to shop. But the relationship is not linear. A 5% increase on a $900 policy barely registers. A 5% increase on a $3,000 policy gets attention. And a 15% increase on any policy almost guarantees shopping behavior. The model accounts for both the percentage increase and the absolute dollar change, along with how the resulting premium compares to market rates for similar risks.

Claims experience is another strong predictor, but not always in the direction you might expect. Policyholders who filed a claim and had a poor experience (slow response, lowball settlement, adversarial handling) are more likely to leave. But policyholders who filed a claim and had a good experience are actually less likely to leave than those who never filed a claim at all. A well-handled claim creates loyalty. A poorly handled one destroys it.

Engagement patterns matter too. Policyholders who never log into their online account, never open emails, and only contact the carrier to pay their bill are more likely to non-renew than those who are actively engaged. Conversely, policyholders who recently called to ask about coverage changes, requested quotes for additional policies, or updated their vehicle information are signaling ongoing engagement.

Life events provide strong but often overlooked signals. A policyholder who recently got married, bought a house, or added a teenage driver is going through a moment where they are re-evaluating their insurance. These moments are both risks (they might shop) and opportunities (they might buy more coverage). Models that incorporate life event data from third-party sources can identify these inflection points.

How the Models Score Risks

Retention models typically produce a score for each policyholder representing their probability of non-renewal at their next renewal date. A score of 0.85 means the model predicts an 85% chance the policyholder will not renew. The scoring is done 90-120 days before the renewal date to give retention teams enough lead time to take action.

The scores are ranked and segmented. The top decile might include policyholders with non-renewal probabilities above 60%. The bottom decile includes policyholders with probabilities below 5% who are almost certain to renew regardless of any intervention. Retention resources are concentrated on the policyholders in the top 2-3 deciles where intervention is most likely to change the outcome.

Critically, the model also estimates the value of retaining each policyholder. A profitable policyholder with a predicted lifetime value of $5,000 is worth more retention investment than an unprofitable one with negative expected lifetime value. Some policyholders that the model predicts will leave are ones the carrier should let go because they are priced below cost.

Retention Interventions

Once flight risks are identified, the carrier needs effective interventions. The most common include proactive outreach from a retention specialist (phone call or personalized email), targeted discount offers on bundled policies, rate-lock guarantees for the upcoming renewal, coverage review consultations that demonstrate the value of the current policy, and loyalty rewards for long-tenured policyholders.

The effectiveness of each intervention varies by policyholder segment. Price-sensitive policyholders respond best to discount offers and competitive rate comparisons. Service-oriented policyholders respond better to personalized outreach and coverage review. The retention strategy should match the intervention to the reason for the predicted departure.

A personal lines carrier that implemented AI-driven retention prediction and targeted intervention reported a 3.2 percentage point improvement in renewal rate in the first year. On their 400,000-policy book, that translated to 12,800 retained policies worth approximately $16 million in annual premium. The retention program cost, including technology, staffing, and discount offers, was approximately $2.5 million.

Feedback Loop

The model improves over time as it observes the outcomes of retention interventions. Policyholders who were scored as high flight risk, received an intervention, and renewed anyway provide data on what works. Policyholders who received an intervention and still left provide data on what does not. This feedback loop allows the carrier to continuously refine both the prediction model and the intervention strategy.

Insurance carriers focused on profitable growth are recognizing that retention is often a more efficient lever than new business acquisition. Acquiring a new policyholder costs 5-7 times more than retaining an existing one, and retained policyholders tend to be more profitable because they have already passed through the initial adverse selection period.