Reprice without waiting for an IT release
Two decisions run an insurer: what to write and at what price, and what to pay when a claim comes in. One is often instant. The other takes weeks and changes as reports arrive.
Underwritingflow
Ratingfactors
Riskscorecards
Claimsevidence
Underwriting and claims
Different tempo, same requirement: a consistent policy and an outcome you can explain.
Underwriting standard risk
Score, price and accept in one request, with external data pulled inside the flow and the policy held as rule sets and decision tables rather than code.
Non-standard and commercial risk
Referrals that need a human view, supporting documents and a surveyor report - a file that stays open and changes as evidence arrives.
Risk pricing
Rating factors, loadings and scorecards as versioned artifacts, with every predictor contribution visible in the explanation.
Claims handling
Triage, fraud signals, assessments, documents and payments - decided across days, with the reasoning attached to the file.
Instant for standard risk, case work for the rest
The split is not underwriting versus claims. It is whether the answer can be reached in one request.
- Real time for accept and price. Standard risks decided in milliseconds through real-time decisioning, with the same policy behind every channel.
- Case work for claims. A claim file where documents, assessments and reports arrive out of order, and a late report re-opens what it invalidated - that is case work.
- Case work for referred underwriting. Commercial and non-standard risks are the same shape as a claim: an open file, evidence over time, and a human view that has to be recorded.
Risk pricing in life and health is high-risk AI
The EU AI Act names risk assessment and pricing in life and health insurance in Annex III, with obligations from 2 December 2027: logging, technical documentation and meaningful human oversight. A policyholder subject to an automated decision can already demand an explanation under GDPR Article 22. The trace answers both.
Questions we get asked
Can we automate part of claims without automating all of it?
Yes, and that is the normal design. Triage, data gathering and rules run automatically; a human review task takes the judgement calls, with the whole file attached.
What happens when a late assessment contradicts an earlier conclusion?
Everything that depended on the old evidence is out of date automatically and re-runs, including re-opening a completed review with the change flagged.
Can underwriting and claims share definitions?
Yes. Both run on the same versioned artifacts, so a risk definition used at underwriting is the same one applied at claim.
How do we evidence pricing decisions?
Every rating decision has a trace showing the factors, the scorecard contributions and the version of the logic that produced the price.
Do we need our actuaries to write code?
No. Rating logic lives in decision tables, scorecards and mathematical formulas that an actuary builds in an editor and can read, test and change directly.