Risk selection & classification
How underwriters decide who gets in and how to group them. All STABLE.
Risk selection
Section titled “Risk selection”Risk selection is choosing which applicants to accept. The underwriter works from:
- Underwriting guidelines — the carrier’s (or MGA’s) rulebook of acceptable risks, defining the risk appetite (which classes are in/out) and referral triggers (which risks must be kicked up for special approval). For an MGA, these guidelines are part of the delegated-authority contract — see 05 The delegated-authority stack.
- Information sources — the application, loss runs (prior claim history), inspections, third-party data/reports, and (for life/health) medical evidence.
The three possible decisions: accept (as applied), accept with modification (higher price, exclusions, lower limits), or decline.
Classification
Section titled “Classification”Classification groups applicants into classes of similar risk so the premium reflects each class’s expected loss. Done well, it is both:
- fair — like risks pay alike; good risks aren’t subsidizing bad ones, and
- accurate — the law of large numbers works within each homogeneous class.
This is the antidote to adverse selection (see 00 Principles of insurance): if you can’t tell good risks from bad, you misprice and the bad ones flood in.
Guardrail: classification must rest on a legitimate, actuarial basis. Grouping that produces unfair discrimination (treating equivalent risks differently without an actuarial reason — or on a prohibited basis) is an unfair trade practice (see 12 Ethics & trade practices). This is exactly the line the AI-bias concerns sit on — see 14 AI insurance regulation.
The novel-risk problem (and the Glacis answer)
Section titled “The novel-risk problem (and the Glacis answer)”Classification needs data. For a brand-new exposure — AI/healthcare Tech E&O — there’s little credible loss history to classify from. That pushes the risk toward:
- judgment rating (pricing from expertise, not tables) — see 02 Rating & pricing, and
- the surplus-lines / E&S market — see 06 Surplus lines / E&S.
Glacis’s contribution: attested evidence of an applicant’s AI governance and controls becomes a new classification signal where none existed. An AI vendor that can prove human-in-the-loop review, bias testing, and model inventory is a measurably better risk than one that can’t — and can be classified (and priced) accordingly. That’s risk selection turned into a product feature.