Where claims, pricing and reserving share their facts

Claims systems exist to settle cases quickly, and I do not ask handlers to work for the reserving team. I ask the system to keep a history that reserving can trust: gross payments, expected and received recoveries, reserve movements and status changes as separate, dated entries. At Arduvelle that ledger feeds the quarterly triangles directly, and the actuaries' view is a documented transformation of it rather than a private extract. I measure quality where it matters to a reserve: the share of motor settlements with a recovery recorded within ten days, the number of manual adjustments actuaries apply each quarter, and the reconciliation of paid amounts to the general ledger. When someone proposes a change to a claims workflow, my team assesses its effect on the triangles before release, and the actuarial function can stop it.

Most data disputes in an insurer are ownership disputes in disguise. Claims operations know what happened on a file; actuaries know how a population of files should be projected; finance knows what was booked. I give each of them ownership of the definitions they understand best and make the connections between them my office's responsibility. Our 340 business terms each have an owner outside my team, recorded in a glossary that the board's audit and risk committee can inspect. Any definition change that would move a reserve, a price or a regulatory figure needs the signature of every owner it touches. This is slower than letting engineers decide in code, and deliberately so. The people who will answer for a number to the actuarial function holder, the auditor or the NBB should be the people who decided what it means.

I treat a model that affects a member's claim or price as a decision with an owner, not as software. Each of the 23 models in our register has a business owner, a statement of what it may and may not decide, monitoring by language, region, age band and occupation class, and a human route around it. Income protection receives the closest attention. It relies on health data, which the GDPR treats as a special category, and its underwriting and pricing fall, as I read the EU AI Act, within the high-risk uses the Act lists for health insurance. Claims triage is not on that list, but we apply the same documentation and oversight, because a delayed payment hurts a self-employed member as much as a wrong price. None of our models declines a claim; models route, and people decide.

Gender has been excluded from insurance pricing across the European Union since December 2012, but fairness questions did not end there. I look for differences the tariff never intended: tenure effects created by years of compounding renewal increases, variables that quietly track age or origin through postcode or occupation, and data gaps that push older policies onto default assumptions. My office does not set prices; the chief actuary does. What we provide is the comparison: for every renewing policy, the price the current tariff would quote a new customer with the same risk, together with the uncertainty wherever data has been imputed. In a cooperative, members are also owners, and the board should see how long-standing members are treated beside the growth figures. I report the renewal gap monthly to pricing and quarterly to the audit and risk committee.

Quarterly Solvency II returns, the annual solvency and financial condition report and the actuarial function's report depend on the same underlying data, and I want them to agree with each other before anyone outside reads them. My reporting-data team runs a controlled cut-off each quarter, and late adjustments after cut-off fell from 52 to 9 a quarter once the claims ledger and the reserving view were reconciled at source. DORA added a different discipline: knowing which ICT providers our critical data flows depend on, how an incident would be classified and who would decide. We built our register of ICT third-party arrangements from the existing data inventory rather than as a separate exercise. With the chief risk officer, I test whether the claims and reporting platforms could be restored within the tolerances the executive committee approved.