The product

Three pillars. One decision layer.

Messy claim in. Governed decisions out. Deterministic gates, agentic assistants, your thresholds at every step.

  1. 01SynthesiseMessy in, structured out
  2. 02EnrichContext at the moment of loss
  3. 03DecideFive gates, one signed record
01
Synthesise

Mess in, structure out.

FNOL arrives in any shape: PDF, photo, voice note, video. Medley extracts, validates and vulnerability-flags before any decision logic runs.

Triage Agent
Predictive routing to the right corridor before human review.
Document Reader
Multimodal document intelligence for receipts, invoices and reports.
Voice Agent
Speech to structured insight, with vulnerability detection.
Prompt guard
Injection scanning on every inbound asset before pipeline entry.
02
Enrich

Add the right context.

Policy at the exact loss date. Endorsements intact. External data through EU-residency-pinned proxies. Degraded-mode fallback so nothing blocks the pipeline.

Policy Assistant
Point-in-time coverage determination with endorsement chain resolution.
Liability Assistant
Fault narrative and third-party liability assessment.
Coverage Assistant
Clause mapping, exclusion testing and endorsement effects at the loss date.
Reserve Manager
Financial exposure modelling and reserve recommendation.
03
Decide

Autonomy earned, not claimed.

Rules at the bottom, living agents at the top. Each gate graduates on its own accuracy, drift and fairness metrics. You sign every level-up.

Atomic decisions
A claim is not one decision — it is 5 on average. Coverage, liability, fraud, reserve, payout. 78% auto-resolve. The hard ones route to humans.
Five maturity levels
Rules → Models → Assisted → Bounded agent → Living agent. Each gate graduates independently. Your risk appetite sets the ceiling.
No model where it matters most
L1 and L2 keep the decision path deterministic. Every model call above that is prompt-pinned, version-locked and replayable.
Carrier-grade decision trail
Every atomic decision signed, replayable and explainable. GDPR Art. 22, EU AI Act, Consumer Duty — by architecture, not bolt-on.
Maturity model

Five levels. You set the ceiling.

The architecture earns the next level. You approve it. Each gate moves independently, so travel flight-delay can run at L4 while liability stays at L2.

L1
Rules
Who decides
Deterministic gates decide
Who reviews
Adjusters review every claim
Models
No model in the decision path
L2
Models
Who decides
Models score and recommend
Who reviews
Adjusters confirm each verdict
Models
Pinned prompts, locked versions
L3
Assisted
Who decides
Auto-resolve above threshold
Who reviews
Adjusters work the exceptions
Models
Replay-tested thresholds
L4
Bounded agent
Who decides
Agents act within authority caps
Who reviews
Humans on the loop, sampled
Models
Canary rollout, auto-rollback
L5
Living agent
Who decides
Agents propose threshold updates
Who reviews
You approve every change
Models
Learning Loop, signed
In the product

The ladder, as your team sees it.

Maturity is a screen, not a slide. Every level, every threshold and every override sits in ClaimOps where the people accountable for them can see and change them.

A decision-maturity control tower in ClaimOps: corridor performance by decision mix, governed STP per corridor, and the typed reasons cases route to review.
01Maturity tower · what each corridor has earned, and what it has not
The governance screen: confidence thresholds per atomic decision, handler override rate, and open governance items awaiting sign-off.
02Governance · thresholds set with the business, and signed
Governance in a second deployment, with its own thresholds, override rates and an EU AI Act replay log.
03The same controls, a different carrier's limits
1 / 3
Under the hood

Three layers wrapping every decision.

Communications, external data integrations and the Learning Loop run beneath all three pillars. Invisible, indispensable.

LAYER · 01

Comms

Tone-aware, deduplicated, paced. A vulnerability override changes cadence and language automatically. Regulator-compliant by default.

LAYER · 02

Data

EU-residency pinned. Region-locked proxies. Every external data fetch creates a timestamped entry in the decision trail. No unbounded model calls.

LAYER · 03

Learning Loop

Closed claims propose threshold updates. You approve every one. The system earns autonomy — it doesn't claim it.

See it on your data.

One corridor, 90 days, production STP. Fixed scope and fee.

See pricing