Insurance application testing

With boundary-hunting: the applicant who turns 25 tomorrow, the postcode on a flood-zone border, the combination of answers that should trigger referral to a human underwriter but quietly doesn't. Quote logic is dense with thresholds, and a wrong price rarely announces itself — it just wins or loses business incorrectly until someone reconciles loss ratios. We map the rating factors and test the edges of each.

Yes, with synthetic claims built to be inconvenient: photos that fail to upload on mobile data, documents in unexpected formats, claims that straddle a policy renewal, duplicate submissions from an impatient claimant. A claims journey is used by people on their worst day, often on a phone, sometimes at a roadside. We test it under those conditions rather than at a desk with fibre broadband.

These are priority flows, because they're where state gets complicated: a vehicle changed mid-term, a cancellation with a refund calculation, a renewal that overlaps an open claim. The demo path — new customer, clean policy, single year — is the least representative flow in insurance software. Real books of business are made of exceptions, and testing has to be too.

Traceable test cases, dated results with environments recorded, and defect reports your compliance function can reference in their own governance. As with our fintech work, we're testers rather than regulatory advisers: FCA and Consumer Duty obligations need their own counsel, and we produce the evidence layer that makes those conversations shorter.

You can't brute-force every combination; a policy with 15 rating factors can produce millions of premium permutations. The answer is risk-based selection: pairwise techniques to cover factor interactions, boundary tests on every band edge for age, sum insured and postcode group, and a regression pack that re-prices a fixed portfolio of reference quotes on every release. If a rate change moves a premium that shouldn't have moved, you find out before your customers or the FCA do.

Document generation is usually a separate system from rating and policy admin, so schedules and endorsements can drift out of sync, and a missing endorsement often only surfaces when a claim is rejected. Testing here means reconciling generated documents against the policy record field by field, regression-testing template changes, and keeping an audit trail of who changed which template and when. In a compliance review, the template history itself is evidence.

Much of the UK insurance stack still runs on decades-old core systems, and institutional knowledge of how they behave has often left the building. The safe approach is characterisation testing: capture the current system's outputs for a broad set of realistic policies before changing anything, then prove the modified system reproduces them. Combined with masked or synthetic test data, this lets you modernise without silently changing anyone's premium or cover.