The Problem
Claims costs on one of the client's largest retail channels were rising faster than the rest of its book. Agreements tied to that channel were a meaningful share of the total, enough that a small move in the volume or severity of those claims moved earnings. Leadership had hypotheses about the cause but no way to tell which mattered, or whether the book had re-rated for good or was reacting to something temporary.
A single trend line could not answer that. Total cost moves for two reasons, more claims or more expensive ones, and a rise in cost per claim can come from the vehicles being covered, from the shops doing the repairs, or from a short-lived problem that will pass on its own. Each points to a different response, from re-pricing the book to managing a handful of providers. The work had to take the number apart before anyone committed to a fix.
The Results
The engagement delivered a diagnosis that turned an ambiguous cost trend into a structured, decision-ready explanation in a few weeks.
The central finding was that the spike was not structural. The book had not re-rated to a permanently higher level of risk. Claims costs on the channel were growing far faster than the rest of the book, and most of that gap came from severity, with volume a smaller part. The severity pressure itself traced to a handful of near-term causes, not a lasting change in the underlying risk. Some of it was a concentration of problems with specific makes and models. Some of it came from the repair side, where costs on those claims ran higher than the vehicles alone would explain.
The deliverable framed the next decision instead of declaring the matter closed. Because the drivers looked concentrated and temporary, the response was less about re-pricing the book and more about isolating the specific models and providers behind the increase. A defined second round would test labor and parts costs, geographic effects, specific component failures, and repair-provider cost patterns, so leadership could act with the structure of the problem already in hand.
Key Techniques
- Data validation against source records before any analysis, with record counts and distributions profiled to confirm the loaded data matched.
- Decomposition of total cost into its two movers, claim volume and claim severity, before comparing channels.
- A mix-versus-rate split of the severity gap, comparing like-for-like vehicles to separate what was sold from what was paid to repair it.
- Separating a structural re-rating from a temporary spike as an explicit question, so the response matched the cause.
- Drill-down to the model and provider level to locate the cost concentration behind the trend.
- Findings read against the book's known seasonal cycle, and all work performed on de-identified data.