Seed
The AI Underwriting File is the only fixed fee in the ladder. It sets the endpoints, the baseline, and the authority limits, and it is deliberately independent: we are paid the same whether the answer is go or no-go.
Regulator-ready agentic AI for carriers
Underwrite your AI.
You price risk for a living. We bring the same discipline to your AI portfolio: one governed workflow, live in weeks on your infrastructure, measured against a baseline your CFO already respects. From the author of Beyond the Prompt.
Self-funding: nothing from your budget. The AI pays for itself from measured value.
Value and Savings Estimator
01The estimate · three levers, five inputs
What would a governed claims agent be worth, where you operate?
Labor cost is not one-size-fits-all, so the estimate starts with geography. Three levers: manual handling cost captured by supervised automation, claims leakage prevented before payment, and a conservative pricing-precision gain from simulation-tested underwriting and selection. Defaults are deliberately conservative and every assumption is visible.
1
02How it is priced · AI always earns itself
Nothing from your budget. Everything from measured value.
Software licenses ask you to pay before the proof. We structure it the other way around: one small fixed engagement to set the terms, then a deployment that pays its own way.
Earn
Deployment is performance-priced: a small slice of each claim the agents actually process, paid out of the value they create. If the workflow does not earn, the meter does not run.
Prove
Someone neutral holds the ruler: baseline before go-live, deltas measured continuously, a kill switch with a named hand. Performance pricing only works when the measurement is independent. That is our job.
Performance terms are structured per deployment with our delivery partners. The review stays fixed-fee and vendor-neutral precisely so the go-or-no-go advice is never contingent on what gets deployed.
03The window · dated, sourced, no panic
Brussels moved the deadline. It did not move the direction.
That is roughly a seventeen-month window in which deploying a governed, supervised agent is calm, cheap, and rewarded. The carriers that wait will do this in 2027 as a compliance fire drill. The ones that move now will do it as an operating advantage.
Now
EIOPA AI governance expectations and DORA already apply to carriers: explainability, human oversight, ICT third-party control.
2 Aug 2026
AI Act Article 50 transparency obligations apply as scheduled; not deferred by the Omnibus agreement.
2 Dec 2027
High-risk obligations for Annex III systems, including life and health risk assessment and pricing, per the Omnibus agreement. Systems already in market benefit from grandfathering.
04Simulation intelligence · rehearse futures, decide deterministically
A decision you can replay is a decision you can defend.
The frontier of insurance AI has moved past reading history. The shift underway is from backward-looking analytics to simulation intelligence: agents trained, tested, and priced in synthetic environments before they ever touch a live claim. Before any engine earns a place in the File, it meets that standard, and the one European supervision is converging on: deterministic where it decides, generative only where it drafts.
Rehearse futures, not just history
History can only price what has already happened. Simulation generates the scenarios your book has never seen — localized climate events, cyber accumulation, supply-chain shocks — so pricing and reserving are tested against futures, not just averages of the past.
Synthetic data, real precision
Mathematically generated data that behaves like your book without being your book: no policyholder data moved, no privacy exposure, and rare events on demand instead of waiting years to observe them. Training and stress-testing scale without the data-wrangling project.
Risk twins before the market
New products, price changes, and portfolio shifts are rehearsed in a synthetic replica of the book first. The launch decision arrives with a simulated loss history attached, not a hope.
Deterministic and replayable where it decides
Where logic touches pricing, coverage, or payment: same inputs, same decision, every time, with the full trace — inputs, rule path, threshold, approver. Generative AI drafts and explains; it does not adjudicate.
The rehearsal happens inside the deployment window, in days, not as a research program. It de-risks the commercial model and satisfies the supervisor at the same time: the performance meter only starts on decisions the simulation has already proven, and the replay becomes the audit trail. It is also where the estimate above understates: rehearsed pricing is what moves the loss ratio, which is why the third lever is deliberately the most conservative number on this page.
05The path · each step sells the next
From a 45-minute briefing to a workflow your board can measure.
The Authority Schedule Briefing
45 minutes, free. You leave with the one workflow you would deploy first and the authority limits you would give it, written in the language your adjusters already live by.
The AI Underwriting File
Fixed fee, four to six weeks, vendor-neutral by contract. Every funded AI initiative scored on six lenses, the regulatory map, the baseline design, and the deployment blueprint.
Deploy and Measure
Your first supervised agent live in weeks on your infrastructure: one region, one line, one claim type, measured against the boring baseline, evidence pack to the board.
If the File does not identify a first deployment you would fund, we will say so in writing.
Book the briefing →06Start here
Bring one AI workflow. Leave with clear authority limits.
Tell us where your AI portfolio stands. We will respond with a focused next step, not a generic transformation pitch.