Move actuarial work off spreadsheets and onto a governed platform.

That something is refused. A production run cannot use uncertified source data unless it is explicitly labelled exploratory — and an exploratory run can never be published, enforced by a database trigger rather than by the route that sets the label. Certification is a decision by a named person, separate from the data-quality run that measures the rows, and an override is matched on exact purpose so a waiver for a pricing study cannot open a reserve close. Alongside that: assumptions versioned with approval and lineage, the approver rejected if they are the author, reproducible runs, and sign-off inside the model lifecycle rather than a binder built afterwards.

Deterministic methods (Chain Ladder, Bornhuetter-Ferguson, Cape Cod, ELR) and stochastic approaches (bootstrap, Mack) for ranges and uncertainty, plus Munich chain ladder to reconcile the paid and incurred ultimates that classical chain ladder leaves apart, and Bühlmann-Straub credibility so a thin segment cannot price itself on its own volatile experience.

Mortality, lapse, persistency and loss-trend studies run continuously against live policy and claims data; approved updates flow into versioned assumption sets that pricing and valuation consume.

Native GMM, VFA and PAA measurement with CSM roll-forward and disclosure outputs, driven from the same governed assumptions and data foundation as pricing and reserving.

Pricing assumptions feed the rating engine, reserving connects to live claims development, and actual-vs-expected monitoring closes the loop — one data fabric across the actuarial control cycle.