Skip to content
Decision-Evidence Operating System

Outcome · Claims cycle time

A claim waits longest where nobody is certain who is allowed to decide it.

The days that go missing are rarely spent working the claim. They are spent in queues, in referrals with no owner, and in the pause between an automated answer and somebody willing to put their name to it.

What the channel costs, measured on the Irish market

0.4yaverage settlement duration through the damage channel — Central Bank of Ireland, NCID mid-year 2025
4.7yaverage settlement duration when the claim litigates — same book, same source, Central Bank of Ireland, NCID mid-year 2025
€20,559average award through the Injuries Resolution Board (median €14,000) — Injuries Resolution Board Annual Report 2025
€88mlegal costs avoided where claims resolved through the board instead of litigating — Injuries Resolution Board Annual Report 2025

How we count this

The Central Bank of Ireland’s National Claims Information Database, private motor, mid-year 2025: average settlement duration by channel — damage 0.4 years, direct settlement 1.8, the Injuries Resolution Board 2.5, litigated 4.7. The spread from 0.4 to 4.7 on one book is a routing outcome, not a complexity outcome — which is why the mechanism below governs the routing itself.The Injuries Resolution Board’s Annual Report 2025: 20,077 applications, more than 10,000 awards (up 17%), €213m awarded in total, and €88m in avoided legal costs. Neither figure is ours; both are printed so you can check them against the source rather than against us.

The same channel durations are load-bearing in our demo-data corpus specification: acceptance gate G-52 requires per-channel settlement duration to match the Central Bank of Ireland channels within ±0.15 years, and the corpus does not ship unless it passes. Specified and gate-carried — a specification the demo must match, never a measurement of the demo.

The pilot metric, and the clause that retires it

Projectedmedian working days from first notice of loss to a sealed decision a third party can re-derive

How we count this

We count the same claim types in both lanes rather than an average across the book, and we count through to the sealed decision rather than to first contact. A lane that answers quickly and reopens is not faster.Two falsifiers, both printed with the target. First: if median cycle time does not improve on the week-0 baseline, the claim is withdrawn in writing — not softened, withdrawn. Second: if cycle time falls while chairs approve more than 80% of verdicts in under 60 seconds, that is rubber-stamping, and it falsifies the result instead of proving it. A target without a falsifier is an advertisement.

The mechanism: the routing decision is itself a decision

Coverage, liability and litigation scores choose the automation depth and the room for each claim — and that choice is written down as an event carrying the autonomy tier it granted, so the routing can be argued about later.

The contested minority get a room whose seats argue from different evidence and a named chair whose ruling is a durable waitpoint rather than an email.

The straight-through lane still passes the gate. The time comes out of the queue, not out of the oversight — which is the only version of this number worth having.

NX/services/nexus-orchestrator/src/routes/dispatch-routes.ts

NX/services/nexus-workflows/src/services/governance.ts

Convening Triage + Graduated Escalation

A meta-decision agent scores each submission or claim and convenes the right room, the right chair and the right autonomy tier — and the routing decision is itself a logged event.

NX/services/nexus-orchestrator/src/routes/dispatch-routes.ts

See it in the console

Exposure map — chat docked
Exposure map — chat dockedSee it in the gallery →

How this lane routes a claim

The tiers carrying inline work and callback work with tenant data are in production. A claim needing accelerated compute, a long-running stateful run, or the verified tier that fails closed routes to a human lane instead, which is slower and honest. The calibration gates above are Irish: only Ireland carries them today, while the GB and Colorado datasets are working sets without acceptance gates of their own.

Run it beside the lane you have

The only honest way to establish this number is a paired baseline: the same claim types, the same period, one lane governed and one lane as it is today. That is how the first pilots are instrumented, and it is why the pilot slot on this page is empty while the market numbers around it are not.

Book the 20 minutes

A person replies with two or three times. Not a sequence.