Parametric triggers

The entry point of the whole actuarial pathway: an insurance payout condition defined on a measurable signal, not an adjusted loss. Everything in this section builds on the missing denominator and the evidence stack in section 18.

What a parametric trigger is

A traditional indemnity policy pays after a loss is reported, investigated, and adjusted. A parametric policy pays when a pre-agreed, objectively measurable condition is met — a hurricane’s wind speed at a named station, rainfall below a threshold, a flight delayed past N hours. No adjuster, no loss history required: the trigger is the contract. STABLE

That last property is the whole reason parametrics lead the Inference Re design: parametric triggers require no loss history and price from attestation data alone. GROUNDED · Brochure

The Inference Re trigger

The draft brochure defines the trigger mechanism for Parametric AI Liability as:

Sustained behavioral drift or scope exceedance beyond a defined threshold, over a rolling 30-day attested window. GROUNDED · Brochure

Unpack each word — each one is load-bearing:

Term	Why it’s there
Sustained	One anomalous inference is noise. A trigger fires on persistence, not spikes.
Behavioral drift / scope exceedance	The two trigger-grade events the Control Platform surfaces against a declared baseline.
Defined threshold	Contractually fixed in the policy wording — no interpretation at claim time.
Rolling 30-day	A moving window, so the trigger reflects current behavior, not a stale assessment.
Attested window	The window is built from notarized receipts, not the vendor’s own reporting — that’s what makes it third-party pricable.

Basis risk — where parametric products go to die

Basis risk is the gap between the trigger firing and the insured actually suffering a loss (or suffering a loss with no trigger). Too much basis risk and the product is either a lottery ticket or a lawsuit. STABLE

This is why vague KPI triggers fail. A Lloyd’s underwriter (call him Mark) rejected a major reinsurer’s KPI-based approach to AI risk in exactly these terms — KPIs and ISO/SOC-style certifications are “no good” for pricing because the distance between “the KPI moved” and “a compensable loss occurred” is unmeasurable. What he endorsed, unprompted, was a specific, countable signal: if a PII filter has been triggered X number of times, “you could probably put a parametric trigger on that.” GROUNDED · KB

The lesson for a producer:

A trigger must be specific (a named signal), countable (a real denominator), and independently attested (not a management assertion).
The Glacis stack emits precisely that shape of signal — filter events, drift measurements, scope-exceedance blocks — each one notarized. The trigger family the underwriter described is the trigger family the platform produces. GROUNDED · KB
If someone offers you a trigger defined on “governance maturity” or “certification status,” you are looking at basis risk wearing a costume.

Why no loss history is (structurally) fine here

The standard objection — “you can’t price without loss history” — inverts the actual situation: nobody has AI loss history, because nobody had a denominator to compute frequency against. Parametric pricing sidesteps this: it prices the probability of the signal crossing the threshold, computed directly from the attested inference population. The signal data exists from day one of an attested deployment. GROUNDED · Brochure

That is also why the denominator’s provenance matters so much — see the trust ladder in 21-03 Insurance signals. A trigger on a vendor-reported count is a KPI in disguise.

→ 01 Claims-verified indemnity

Drill this page →17 bank questions stand behind what you just read. Check it while it’s warm.