Lesson 05 · Regulated · 12 min
Insurance: consistency as the product
Scope
Educational, not legal or compliance advice. Insurance conduct rules vary by country and by US state; treat the references as pointers to the primary sources for your market.
Claims and pricing have an unusual property: consistency matters more than sophistication. Two comparable claims routed differently is a fairness problem, a regulatory problem, and a complaint — even when both routings are individually defensible. In the EU, insurance risk assessment and pricing is a named high-risk use; in the US, unfair-claims-settlement-practices laws bear on inconsistent handling. Either way the pressure is toward a fixed, stated, repeatable decision path.
Sources (2) — primary references, with jurisdiction and verification dates
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Annex III — European Union
Classifies as 'high-risk' the AI uses this course touches: creditworthiness scoring, risk assessment and pricing for life and health insurance, and CV screening / recruitment. High-risk obligations (logging, human oversight, accuracy) become enforceable 2026-08-02.
European Union / EEA · published 2024-08-01 · last verified 2026-08-28
- Regulation (EU) 2016/679 (GDPR), Article 22 — automated individual decision-making — European Union
Gives individuals the right not to be subject to a decision based solely on automated processing that significantly affects them (e.g. automatic refusal of credit, e-recruitment), and a right to human intervention. This is the legal backbone under a required human checkpoint.
European Union / EEA · published 2016-04-27 · last verified 2026-08-28
That pushes hard towards structure. The decision path is fixed, the routing rule is stated in the output, and the same inputs produce the same route every time. Sophistication would be actively harmful here, because it introduces variance.
- State the rule that produced the routing. A route without a rule cannot be checked for consistency.
- Test with near-identical pairs. Vary one field and confirm the route only changes when it should.
- Route; do not decide. A claim decision carries appeal rights that attach to a person.
- Watch for drift. The same model on the same input can route differently after a version change.
The lapsed-policy case is the trap. Everything about the claim reads as routine, and the only disqualifying fact is a date comparison the agent must actually perform rather than assume. An agent that pattern-matches on "collision claim, moderate value" will fast-track it.
The evals course is the tool here. Build pairs of near-identical claims, assert that the routing matches, and run them on every model change. Consistency is testable in exactly the way that judgement is not.
Checkpoint
Produce a run that catches the lapsed policy and names the rule it applied.