Strategy · Insurance claims

The carriers bought the workflow, not the model.

Top-five US insurers moved AI into production claims work this quarter. None of them built a model. They bought decisions inside a system they already ran, and kept the regulatory exposure the contract cannot move.

Field briefing · HB-2026-0018 The Hikari Blue team · Operator notes from Austin 5 min read

A top-five US insurer put AI subrogation into production this quarter. It did not build a model. It bought a decision inside a system it had already run for years.

The enterprise conversation is still framed as build versus buy at the model layer. In the American claims stack, that question closed while the debate continued. The AI that reached production scale in the second quarter arrived embedded in the system of record, sold by the vendor that already held the carrier's data, the repairer network and the parts catalogue.

What the quarter actually showed

CCC Intelligent Solutions, whose network sits between carriers, repair shops and suppliers in US auto claims, reported second-quarter revenue of $285.9 million, up 9.8% year over year, with adjusted EBITDA of $115.5 million (CCC Intelligent Solutions, July 30, 2026). Management attributed four points of that growth to the products it classifies as AI-based. Those products now represent roughly 11% of revenue, more than $120 million annualized, growing near 45% to 50% year over year, and net dollar retention came in at 107%, up from 106% for full-year 2025 (CCC Q2 2026 earnings call, July 30, 2026).

The named deals carry more signal than the totals. One top-five insurer became the largest carrier to date to adopt the AI-powered subrogation product, with deployment scaling under a multi-year agreement. Two other top-five insurers extended AI-enabled claims workflows by deploying FirstLook, the early total-loss identification product (CCC Intelligent Solutions, July 30, 2026). On the call, management described carriers signing after extensive testing, wanting proof of value before committing.

That is the change. Not that AI works in claims, which vendors have asserted for three years. That buyers of this size in a regulated line finished testing and signed multi-year production commitments in the same quarter. The disclosure names the tier, not the carriers, so whether the three deals sit with two firms or three is not public.

Where the value is created, and who books it

Three workflows, three distinct mechanisms. Subrogation is recovery: money returned from the at-fault party, a revenue line, not a cost line. Early total-loss identification is cycle time: a file called correctly on day two instead of day twelve stops paying for storage, rental and adjuster hours on a vehicle that was never going to be repaired. Claims routing is allocation: the file reaches the handler who can close it, the first time.

Each of those mechanisms lands in loss adjustment expense or in recovered dollars. Neither shows up in the public record. The vendor's gain is disclosed quarterly and priced into a renewal. The carrier's gain is disclosed nowhere.

That asymmetry is the first executive question. Net dollar retention of 107% means the installed base pays more this year than last, and the AI line is the fastest-growing part of it. A carrier's recovery uplift and expense reduction have to clear the price of that line before a single point of margin reaches the underwriting result. Nothing published this quarter says by how much.

What does not transfer with the contract

The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers on December 4, 2023. As of April 1, 2026, twenty-five jurisdictions have adopted it, twenty-four states and the District of Columbia, with California, Colorado, New York and Texas running their own insurance-specific AI guidance alongside (NAIC, 2026).

Two provisions decide how a bought workflow is treated. Section 1.8 requires the insurer's written AI Systems Program to cover AI used in regulated insurance practices whether developed by the Insurer or a third-party vendor. Section 4.1 requires due diligence sufficient to ensure that decisions made or supported by a third-party AI system, where they could lead to adverse consumer outcomes, will meet the legal standards imposed on the Insurer itself (NAIC Model Bulletin, 2023).

You can buy the workflow. You cannot buy the answer you owe the examiner. That stays on your side of the contract, whoever trained the model. Hikari Blue · operator note

Section 4.2 names the two contract terms that make this operable: audit rights, or audit reports from qualified auditing entities, and a duty on the third party to cooperate with the insurer on regulatory inquiries and investigations related to the insurer's use of that product. A subrogation referral or an early total-loss call is a decision affecting a consumer. Under Section 4 of the bulletin, a department can ask what the system did, on what data, under whose oversight. Many AI line items were added to master agreements signed before the bulletin existed, and those agreements carry neither clause.

The direction of travel is set. The NAIC formed a Third-Party Data and Models Working Group in 2024 to build a regulatory framework around third-party AI data and models used by insurers, work still under way (NAIC, 2026). The framework is not finished. The exposure is already live in twenty-five jurisdictions.

What separates the leaders

It is not who has AI in claims. Three production deals at the top of the market in one quarter say that stops being a differentiator. The separation is between carriers who wrote the evidence terms into the agreement at signature and carriers who negotiate them after a market conduct examination has already asked the question.

Three things an executive committee can examine this month.

  • Which vendor-embedded AI decisions in the claims flow touch a consumer outcome, and whether they sit inside the AIS Program scope or outside it because no one labelled a routing rule as AI.
  • Whether the carrier can produce, without calling the vendor, the input, the model version and the human override for one contested subrogation referral and one early total-loss call.
  • What the AI line costs at renewal, and what recovery uplift and loss adjustment expense reduction it has to clear to be worth the accountability that comes with it.

Two metrics carry the answer over the next four quarters. Recovered dollars per closed claim and loss adjustment expense per claim, tracked against the AI line item in the vendor contract rather than against last year's baseline. And the share of vendor-supplied AI decisions the carrier can evidence unaided. The first tells you whether the economics are yours. The second tells you whether the exam is survivable.

The question to bring to the next executive committee

Which decisions in your claims workflow can you evidence without calling your vendor?

Buying the workflow was the right call. Buying it without the record is how a productivity gain turns into a regulatory finding.

The Hikari Blue team · Austin, July 2026

More from the Newsroom

See all articles in the Newsroom →

Own the record on workflows you did not build.

Thirty minutes with an operator. No slides.

Direct call with one of the partners. We listen, we structure, and we tell you what an evidence layer over vendor-supplied AI decisions (scope, contract terms, audit trail) would look like for your specific stack and regulatory perimeter.

Book a 30-minute working session