Sharper claim forecasts, half the idle reserve.

The problem

Every open insurance claim needs a cost estimate, and the sum of those estimates decides how much capital has to be held in reserve. Estimates were rough, so reserves were set conservatively, and a lot of capital sat idle, covering uncertainty rather than real risk.

Our approach

• A two-stage forecasting model first predicts whether and how a claim will escalate, then predicts its cost, which handles the many small claims and the rare expensive ones far better than a single model.
• A Monte Carlo reserve simulation combined claim-level forecasts and their uncertainty into a distribution of total portfolio cost, so the reserve could be set at an explicit confidence level instead of a rule of thumb.
• Clear outputs for finance teams: the distribution, the chosen confidence level and the capital it implies, side by side with the current reserve.

The result

Claim-level cost-forecast error fell by 69%. The portfolio simulation showed that 49% of the over-committed reserve capital could be released while still covering the same level of risk.