National Auto Carrier

A national auto carrier managing approximately 80,000 policies faced a core profitability challenge common in personal lines P&C: loss costs vary dramatically at the individual driver level, yet traditional pricing models aggregate risk in ways that obscure that variance. The carrier could not reliably distinguish its most profitable customers from its least profitable ones without completing a full underwriting application — far too late in the acquisition funnel to redirect marketing spend or divert high-risk applicants. Without granular, early-funnel loss cost intelligence, the carrier was acquiring unprofitable policyholders at full marketing cost, eroding loss ratios and limiting the effectiveness of its digital lead strategy.

The carrier engaged Pinpoint Predictive, an AI firm purpose-built for insurers, to deploy a Loss Cost Prediction model trained on historical policy and claims data. The model's key design constraint was top-of-funnel operability: it generates a risk score using only a name and address, meaning the carrier could apply it before a prospect completes a full application — before underwriting resources are consumed. The predictive ML model segments applicants by predicted loss cost decile, enabling two distinct interventions: routing the highest-risk applicants to alternate application journeys, and concentrating marketing budget on the lowest-risk, highest-profit segments. Pinpoint's ROI Calculator was used to project financial outcomes by applying the model's lift-by-decile to the carrier's existing book of business, providing a transparent, auditable forecast prior to deployment.

In year one, the model delivered a 4.61-point loss ratio improvement — worth over $5 million — despite the book shrinking by a projected 7% as unprofitable segments were deprioritized. The 3-year projected ROI exceeds $19 million, with a cumulative loss ratio improvement of 7.37 points. The model's discriminatory power was validated through lift analysis:

• Highest-risk 5% of predicted drivers: actual loss ratio 1.6x above average
• Lowest-risk 5%: actual loss ratio 1.3x below average
• Top-to-bottom loss cost spread: 4.5x ($264 vs. $1,177 actual loss cost)

These results demonstrate that the model's risk segmentation meaningfully exceeded the carrier's existing baseline pricing in identifying outlier risk.

• Early-funnel scoring with minimal data (name + address) allows carriers to act on risk intelligence before underwriting costs are incurred — maximizing the leverage of each intervention.
• A shrinking book is not a failed outcome: systematically shedding unprofitable segments can generate more ROI than growing a poorly segmented one.
• Lift-by-decile analysis should be the validation standard — top/bottom spread quantifies model value in terms underwriters, actuaries, and executives can all act on.
• Carriers should distinguish between rate adequacy and segment selection; even a well-priced book contains profitable growth opportunities hidden within existing pricing tiers.
• Projecting ROI using the model's actual lift deciles — rather than average improvement assumptions — produces credible business cases that survive actuarial scrutiny.