An actuarial workbench for life insurance: mortality experience studies, claim analysis and IFRS 17 projections in one web app, with an AI assistant that answers by text or voice.
Experience studies, claim reviews and IFRS 17 runs follow the same steps every time: pull the policy and claim data, clean it, map products, compute exposure, compare against a table, then cut the result a dozen ways for the report.
Done in spreadsheets, each run is slow, hard to repeat exactly and hard to review. ActuarialPro turns those steps into guided screens with the method written down once, so a study can be rerun, compared and exported the same way every time.
Actual-to-expected deaths on age last or nearest birthday, with credibility, broken down by age band, gender, product, product group, policy year, calendar year, sum insured and cohort. Studies are saved and can be rerun from a snapshot.
Claim ratio, incidence and policy-year experience, early-claim review, segment rankings, and anomaly flags for claims worth a second look, down to claim-level detail.
Monthly cash flows per group of contracts, the LRC with best estimate, risk adjustment and CSM, accounting movements, and a reconciliation back to the source workbook.
Answers by text or voice, researches public sources and checks the excerpts it quotes, cites up to five sources, writes equations properly and can open the right screen in the app.
One place to manage mortality, lapse, discount-rate and illiquidity-premium tables that every tool reads from.
Extraction straight from SQL Server or upload from Excel, a database explorer, role-based access and CSRF protection.
The app is Flask with server-rendered pages and plain JavaScript, a small SQL database for users, saved studies and tables, and a direct SQL Server connection for extracting policy data. Long calculations run as background jobs, so a big study never freezes the page, and money is summed in integer cents to avoid rounding drift.
The IFRS 17 projections live in a separate engine package written with NumPy and pandas. It can be tested and run on its own, and ActuarialPro calls it as a library from the IFRS 17 page. When a reference workbook is supplied, each run is reconciled against it, and anything that needs a human check is listed as a review issue rather than hidden.
Benchmark figures are from the engine’s own synthetic test data on a local machine, not production.
All screens use dummy policies, claims and tables generated for this page. No company data is shown.