Home / Where we work / Insurance

Insurance.

AI & data consulting for life, annuity & pension risk transfer platforms · reinsurers, claims and specialty carriers where the same file shows up

A 400-page file lands. Cycle time is set by how fast one person can read it, and the price is set by an assumption nobody has refreshed since last year.

Why it takes days

Two problems wear every costume in insurance: reading, and assumptions.

Reading is capped by people. Records read at the speed of one underwriter, one adjuster, one analyst on the cleanse. Throughput is a headcount number.

Assumptions are stale by design. A study takes a quarter. By the time the assumption lands in pricing, the block has moved. Acquired blocks carry the seller’s assumptions until someone gets around to reviewing them.

Both answer to three rooms. The chief actuary, the auditor and the regulator. Twenty-five states and DC have adopted the NAIC AI bulletin, and Bermuda scales the BMA every year. Documentation that satisfies the modeler and nobody else is the most common failure we see.

Already running an accelerated program? Then you already know the constraint is the sign-off, not the model. We build to the sign-off first.

What we build

Agents that read. Assumptions that refresh. Documentation that clears the room.

What the leaders are doing

The standard is changing. Three examples from the last eighteen months.

John Hancock · Sept 2025

John Hancock and Munich Re doubled or tripled instant-decision rates, now up to $5M of face.

Carriers on the expanded alitheia platform are “often doubling or tripling” prior instant-decision rates, and John Hancock extended instant eligibility to $5M (company release). The reading problem, solved at the top of the market. Expect distribution to ask why your rate is where it is.

CCC · Dec 2024

CCC paid $730M for EvolutionIQ. Seven of the top fifteen US disability carriers already used it.

The market has decided that reading claim files is a machine’s job (company release). What is still open is who turns the reading into better reserving.

LIMRA · March 2026

US pension risk transfer was $48.8B in 2025, the third-strongest year on record.

Buy-ins hit a record $17.5B. At that volume the desks that quote fastest on the cleanest data take share, and the ones running the cleanse as a manual project pad the quote or miss the window.

Two to four weeks against your own accuracy bar. Fixed fee from $15,000; worth ten times the fee or you do not pay it.

Book 30 minutes →
The first step

Two to four weeks. Measure before anyone commits to a build.

  1. Pick one file typePhysician statements, scheme data, bordereaux, claim files. Whichever hurts most.
  2. Measure against your barAgents run on a sample of your files, scored against your golden set and your reviewers’ decisions. You see the eval, row by row.
  3. DecideA number, a costed path, and where the days go on a PRT bid. Or a reason not to build yet.
Fixed fee, from $15,000. If it does not show a path worth ten times the fee, you do not pay it.Our founder is a Fellow of the Society of Actuaries who ran the technical risk review that cleared John Hancock’s accelerated underwriting model, led the experience studies behind $754M in reserve accuracy at Manulife, and built the program that digitized 50M+ pages of medical records at 99.9%. Start with the briefing and diagnostic, or go straight to a Document-to-Decision sprint. The work that follows is the Document-to-Decision sprint on the file type you chose, with Data Foundations underneath it.
Does this sound like you?

Different file, same two problems.

Instant-decision rate stuck. Cycle time set by reading speed. Experience studies stale on arrival.

Every bid a data project on a deal clock; every acquired block a migration fire drill. Roughly 95% of the fields overlap from one pension risk transfer deal to the next, and it is still done by hand.

Bordereaux as Excel from one cedant and text files from another. One policy across three quota-share sessions. The BMA checklist the same every year and answered from scratch.

Hundreds of adjusters each holding hundreds of files of 500 to 50,000 pages, and a reserve set without reading them.

Submissions arriving faster than underwriters can open them, good risks unread, and no hit-ratio data because the submission was never structured. More on specialty and E&S.

Common questions

Asked on most first calls.

What does the diagnostic cost?

Fixed fee, from $15,000. If it does not show a path worth ten times the fee, you do not pay it.

Why do PRT quotes take three weeks?

The pricing rarely takes the time; the data cleanse does. Missing dates of birth, stale vital status, benefit definitions that vary by scheme. When the cleanse runs late, the quote gets padded, and padding loses bids. A repeatable cleanse puts the days back on pricing.

Can agents really read a 400-page underwriting file?

Reading is the part that works today: attending physician statements, labs, applications, mapped to findings with page references. The part that needs care is what happens after the reading, which is why every finding routes to the underwriter and instant offers issue only where the rules, not a model, say the case is clean.

We answer to a regulator and a rating agency. How does AI fit?

Everything ships with the file an examiner expects: evals before production, an action log, recorded human sign-off, and documentation written for the committee that has to defend it. Built by a Fellow of the Society of Actuaries.

Measure before you build.

Two to four weeks. One file type, run against your own accuracy bar, or your last PRT deal run through the cleanse. Fixed fee, from $15,000. If it does not show a path worth ten times the fee, you do not pay it.

If it’s not a fit, we’ll tell you who is.