78% of P&C insurers have adopted generative AI in claims. Only 4% have scaled it meaningfully. What follows is a working map of who's actually shipping, what's working, and where the industry's structural constraints are shaping where AI can land.
The AI claims market split hard in 2025 and the split is sharper now. The companies getting traction all share some combination of these four traits. The companies struggling usually lack at least two.
Multi-agent orchestration beats single-model approaches. The winners all run specialized agents across intake, triage, evaluation, correspondence, and routing. No one credible is shipping a single model that does everything.
Narrow scope beats platform scope. Kyber built a business on claims correspondence alone. Avallon does back-office automation. The narrower the initial wedge, the clearer the ROI story and the faster the deployment.
AI-native architectures beat bolted-on AI. ClaimSorted and Elysian own the workflow end to end. They're not integrating AI into an existing TPA process; they built the TPA around AI from day one. That architectural choice shows up in every metric.
Domain credibility still matters. Grace Hanson is a five-time Chief Claims Officer. ClaimSorted pulled senior talent from Hiscox, Lemonade, AXA, Hartford, Liberty Mutual. Teams without insurance DNA are not closing large deals.
Five Sigma's Clive is the clearest win in the category. It's a multi-agent AI claims expert trained on a decade of adjuster practices, deployed as an augmentation layer across Guidewire, Duck Creek, Sapiens, and direct-to-carrier stacks. The 2025 Technology Partner of the Year award wasn't hype; the customer momentum followed.
Q1 2026 added Starr (global specialty and investment carrier) and Loadsure (freight MGA) as named production deployments. Both chose Clive specifically because it layers onto what they already run. Five Sigma isn't asking carriers to rip out ClaimCenter or PolicyCenter. That's the whole strategy.
One of the largest InsurTech seed rounds on record. Pavel Gertsberg and German Mikulski (previously built pet insurance company Fluffy to 20K+ customers) went into YC S24 with a TPA thesis and came out with Atomico leading the round. The team is recruited from Hiscox, Lemonade, AXA, Hartford, and Liberty Mutual.
The architecture is the differentiator. ClaimSorted isn't a TPA using AI; it's an AI platform operating as a TPA. Commercial and personal property, auto, small commercial, general liability, accident and health, warranty. Tens of thousands of policyholders served since launch across the US, UK, and EU. Claim leakage below 1.2% and reopen rates below 1% within 30 days put them on a different performance curve than legacy operators.
Avallon is the clearest example of the narrow-wedge thesis. It doesn't compete with Origami, Kinetic, or any claims platform. It plugs into them. Voice AI for claim intake and status calls (Twilio, Vapi), document processing, structured data entry. Athens Administrators, a California workers' comp and multiline TPA, uses it to auto-resolve 64% of inbound calls.
Founded by Cornell and MIT grads, including a founding US engineer from FINN. The stack is intentionally modern and integration-first: React, Node, PostgreSQL, AWS. The pitch to a TPA is simple: you're chronically short-staffed, your intake team is drowning, here's an agent layer that handles intake, status, and billing calls without replacing your CMS.
Elysian's differentiator is Grace Hanson's operating resume: Chief Claims Officer at AIG, Allied World, Homesite, Hiscox, and Hippo. That kind of trajectory isn't window dressing at a startup, it's the entire go-to-market. Commercial P&C buyers have to trust the claims expertise before they'll trust the AI.
March 2026 brought the Lloyd's Lab Cohort 16 selection: 12 companies chosen from 220 global applications. During the program, Elysian is adapting its Claim Conductor platform for Lloyd's specialty lines with real-time claims quality review, coverholder and TPA oversight, and syndicate compliance reporting. That positions Elysian as the claims AI story for a market segment (Lloyd's specialty) that no other AI entrant is credibly serving.
Kyber is the case study for narrow-wedge wins. Claims correspondence only. Status letters, coverage determinations, denial explanations, third-party acknowledgements. High volume, legally sensitive, compliance-heavy, and repetitive enough that the ROI math is obvious within a quarter.
November 2025 brought the Guidewire Accelerator for ClaimCenter, shrinking deployment time from months to weeks. February 2026 added the Majesco Intelligent Claims integration, which gives them bi-directional coverage of two of the three biggest P&C policy platforms. Aspire General automated 44% of outbound correspondence within 90 days of go-live, with what had been 3-5 minute human tasks now handled end-to-end instantly.
Shift was the fraud detection leader. In September 2025 they launched Shift Claims, a full agentic AI platform that assesses and prioritizes cases, guides adjusters, and automates tasks across the lifecycle. The strategic read: fraud detection was a wedge into carrier workflows, and they're cashing in the access.
The AXA relationship is the proof point. The two renewed a five-year strategic partnership in March 2026, spanning 15 countries across Europe, Asia, and Latin America. AXA Switzerland is the early production adopter for Shift Claims. Early rollout metrics show 3% reduction in claims losses, 30% faster handling, 60% automation rate, >99% accuracy in assessment.
Guidewire's Olos release (December 2025) pulled a lot of strategy into one update: Underwriting Assistant (agentic GenAI), expanded workers' comp claim segmentation, GenAI Service, Agentic Framework, Claims Intel integration. The acquisition of ProNavigator closed in March 2026, adding a knowledge management layer that surfaces institutional context inside workflows. ProNavigator had 34 customers, 12 of them already shared with Guidewire.
The strategic play: Guidewire wants to be the platform where AI happens, not a platform that AI happens around. That's a direct response to Five Sigma, Kyber, and every ISV building AI layers on top of ClaimCenter. The Accelerator program is how they're trying to keep partners in the ecosystem instead of losing them.
The rebrand was more than a logo change. Cotality repositioned from mortgage and property data toward "property ecosystem intelligence" for insurance, with claims, risk, and underwriting in the same story. The Claims Workspace suite (Claims Workspace, Claims Estimate, Claims Scope for AI roof analysis, Claims Validate for automated QA) is the foundation. The Digital Hub Alliance adds Itel for materials analysis, Yembo for virtual inspections, ION for field intelligence, LexisNexis Flyreel for risk assessment, Baker Tilly for restoration advisory, and Hancock Claims Consultants for lifecycle support.
The API story is where Cotality diverges from Verisk. Published REST APIs. Developer Portal. Stated architectural commitment to partner integrations. That's the posture of a platform that wants to be built on, not a closed ecosystem that rents access.
Almost every claims AI decision in property eventually hits this question. The answer shapes what's possible.
| Verisk ecosystem | Cotality ecosystem | |
|---|---|---|
| Market position | ~90% penetration in property claims routing via XactAnalysis. Xactimate is the estimating standard. | Growing challenger. Strong in property data, expanding into claims and underwriting intelligence. |
| Architecture posture | Closed ecosystem. Gated API access. Integration partners are curated and limited. | Open API. Published Developer Portal. Digital Hub Alliance of integration partners. |
| AI direction | XactAI (Sept 2025). Adjuster productivity features inside existing products: note summarization, photo labeling, transcription, ALE receipt categorization. | AI woven into product roadmap (Claims Scope for roof analysis, Claims Validate for QA) with ecosystem extensibility. |
| Integration cost | High. API access negotiated, priced, and limited. | Lower. Self-service Developer Portal, published REST APIs. |
| Startup activity | Funded AI startups consistently build adjacent to, not through, Verisk. | Actively courting partners. Itel, Yembo, ION, Flyreel in the Alliance. |
| Buyer implication | If operations run through Xact, AI opportunities are narrower and more constrained by ecosystem rules. | More room to design an AI workflow that the vendor doesn't gate. |
This isn't a moral judgment on Verisk. XactAnalysis is the market standard because it works, and XactAI is a reasonable productivity response to the industry's AI pressure. But the architectural posture shapes the space of possible AI investments. A carrier running through XactAnalysis has less room to bring in a Five Sigma or an Avallon and get creative with the workflow. A carrier that's Cotality-aligned, or that operates outside both, has more freedom to choose AI based on fit.
The sharpest read: well-funded AI claims startups have made the same call. They're building adjacent to Verisk, not through it. That's a data point, not a theory.
These are the calls I'd walk into any claims executive's office with, given the current map.
Any vendor pitching a single model doing everything should be evaluated skeptically. The pattern is specialized agents handling specific workflow stages. If the architecture story isn't that, ask why.
Correspondence. Intake triage. Estimate QA. The wins that scale are the ones where the scope is tight enough that the ROI math is obvious after 90 days. Platform ambitions come later.
If your operations are locked into XactAnalysis, the AI conversation is different. Knowing where the gates are changes what you evaluate. If you have Cotality or hybrid architecture, you have more room to move.
The gap isn't a technology gap. It's an implementation and measurement gap. Every serious engagement should start with a scorecard: accuracy, failure modes, comparison to current process. "It seems pretty good" is not a decision.
The AI-native TPAs winning business are the ones that pulled senior claims people into the build. If the vendor team is technology without insurance credibility, or insurance credibility without technical depth, that gap will show up in the deployment.
Guidewire's Olos and ProNavigator acquisition are real. Cotality's AI story is real. But neither is shipping at the pace of specialist AI startups. If native is the only option you consider, your roadmap slows to the platform's cadence.
I help claims leaders translate this landscape into specific build-or-buy calls for their operation. Two-week sprints, real data, measured results. Initial conversations are always free.
Start a conversation ↗