Insurance · Deep dive

Tractable

London-born computer-vision company whose AI reads photos of crashed cars and damaged homes and writes the repair estimate — the world's first computer-vision unicorn in financial services (June 2021), ~$185M raised from Insight, Georgian and SoftBank, $7B in claims processed annually through its platform, customers including GEICO, The Hartford, Aviva, Admiral and all four of Japan's largest insurers — now three years past its last round, past a founder-CEO handoff, and past a seven-year legal war with CCC that ended in a quiet January 2025 settlement.

emerging

The question that decides it: Tractable sells AI that writes damage estimates, but the estimate itself lives inside someone else's system of record — CCC holds roughly 85% of US auto estimatics by Tractable's own March 2024 antitrust filing, and CCC, Solera (Qapter) and Mitchell all now bundle their own photo-AI into the platforms insurers already pay for, with CCC alone booking on the order of $100M in AI revenue. Can an independent estimating layer keep winning carrier contracts on accuracy and touchless-rate superiority — as it did with GEICO, The Hartford and all four top Japanese insurers — when the January 2025 CCC settlement ended its legal challenge to that gatekeeping, no new capital has arrived since July 2023, headcount is down roughly 40% from peak, and multimodal foundation models are commoditizing the computer vision that was the moat?

Snapshot

Tractable trains computer vision on insurers’ decades of claims photos and repair invoices so that a policyholder’s smartphone pictures come back, in minutes, as a costed repair estimate. The archetype of applied AI before the LLM era: founded at Entrepreneur First in 2014, ~$185 million raised, a $1 billion Series D in June 2021 — the first computer-vision unicorn in financial services — and roughly $7 billion in claims value flowing through the platform annually by mid-2023, for customers including GEICO, The Hartford, Aviva, Admiral and all four of Japan’s largest insurers. It matters now as the cleanest test of the question this site’s same-day CCC teardown poses from the other side: can an independent AI estimating layer survive once the systems of record bundle their own? Tractable enters that test with no new capital since July 2023, headcount down roughly 40% from peak, a new operator CEO since September 2024, and a freshly settled seven-year legal war with CCC.

Founding story

Alex Dalyac was a French Imperial College machine-learning graduate with about a year of work experience when he joined Entrepreneur First’s London cohort in 2014, convinced that post-2012 deep learning could automate expert visual judgment. At an EF hackathon he met Razvan Ranca, a Romanian ML researcher just out of his master’s; they moved in together, spent months on customer discovery, and picked auto damage appraisal for a coldly practical reason: insurers sat on enormous archives of claims photos paired with human-written estimates and repair invoices — labeled training data nobody was using (Entrepreneur First; Dalyac’s TechCrunch essay, July 2021). Adrien Cohen, a co-founder of Rocket Internet’s Lazada, joined as third co-founder after the seed round in 2015 and built the enterprise sales machine. The founder era ended in stages — Dalyac ran the company for ten years, then moved to founder-chairman in September 2024, handing the CEO role to software product executive Venkat Sathyamurthy; Dalyac’s own attention has since shifted to a new venture, YouSquared AI (LinkedIn, 2025-26). Ranca remains the technical continuity.

How it works

Mechanically, Tractable inserts itself between first notice of loss and the estimate. After an accident, the insurer’s FNOL flow sends the policyholder a link; a guided web capture walks them around the car, prompting for corner shots, close-ups and VIN. The images hit vision models trained on millions of historical claims photos joined to the estimates and invoices that resolved them; the models identify each part, grade severity, and make the appraiser’s core call panel by panel: repair, blend, or replace. The output is not a verdict but estimate lines, priced against local labor rates and parts data and written into whatever estimating platform the carrier runs — CCC, Mitchell, Audatex, or Verisk’s Xactimate on the property side. Straight-through claims settle without a human touching them; exceptions route to human appraisers with the AI’s work pre-populated. The same engine runs in reverse as AI Review, auditing repair-shop estimates against the carrier’s standards, as AI Triage, splitting repairable vehicles from probable total losses at FNOL so salvage starts sooner, and as AI Subro, re-checking the photo evidence in other insurers’ subrogation demand packets. The hard limit is physics: the AI prices only what a camera can see. Suspension damage, bent frames and flood-soaked interiors are invisible to it — which is both the source of the body-shop criticism below and the reason humans stay in the loop. Spain’s Admiral Seguros showed the ceiling when the flow works: 12,000 touchless claims in 2021, 90% of estimates produced without a human appraiser, 98% completed inside 15 minutes (Tractable case study).

Product and business overview

The auto claims suite — AI Estimate, AI Review, AI Triage, AI Subro — is the revenue core, sold to carriers and integrated through the estimatics rails and core-systems ecosystems (Guidewire PartnerConnect, January 2022; Duck Creek, April 2022; Mitchell’s Intelligent Estimating used Tractable’s vision from 2020). The property product, launched 2022, applies the same pipeline to homes — policyholder photos in, damage identified, classified and measured, estimates generated on Verisk’s Xactimate under a February 2023 partnership. A third leg, the auto ecosystem business, sells condition assessment outside claims — parts distributors, recyclers and used-vehicle players — and was reportedly growing volumes ~10x year over year in 2023, as was Japanese property (TechCrunch, July 2023). An applied-science team has kept the research posture alive, and under Sathyamurthy the 2025-26 positioning has shifted to agentic claims orchestration — AI as the workflow layer, not just the estimate (SoftBank Vision Fund interview; Everest Group Top 50 P&C tech providers, 2026).

Business model and pricing

Revenue is enterprise contracts with usage economics: carriers pay per assessment/claim processed under annual or multi-year agreements, with no published rate card — every public path ends in a sales conversation, normal for a vendor whose deals are seven-figure carrier integrations. The company’s preferred metric, claims value processed (~$7B annually, July 2023), measures customers’ repair spend, not Tractable’s cut. What is known of the P&L comes from Dalyac’s candid July 2023 interview: SaaS-grade margins, still unprofitable, nearly at EBITDA breakeven after cost cuts he described as brutal but necessary post-ZIRP (TechCrunch). GetLatka pegs ARR at $79.4M (2024) — an unverified scrape, but directionally consistent with a company that reported 600% revenue growth over the 24 months to June 2021 and then stopped publishing growth numbers.

Traction over time

Two things jump out. The 2019-21 run was genuinely explosive — 600% revenue growth, GEICO and the entire Japanese market landed inside 18 months. Then the public numbers freeze: the $7B claims figure is the same in July 2023 as in June 2021, headcount fell by roughly a third or more, and no customer, revenue or valuation milestone has been announced in the three years since the Series E (through July 2026). Growth may have continued privately; the disclosure silence is itself a data point.

Market analysis

The spend Tractable automates is loss-adjustment expense — the human cost of appraising claims — sitting atop hundreds of billions of dollars in annual auto and property repair payouts across its markets. The software wrapper: claims processing software globally was estimated at ~$40.8B in 2024, growing ~8% (Grand View via Verified Market Reports); the AI-native claims niche is sized far smaller but faster — under $1B by 2030 at ~16% CAGR by one estimate (ResearchAndMarkets, 2026). The structural tailwinds are real: an aging, shrinking adjuster workforce (acute in Japan, Tractable’s co-largest market), photo-first FNOL normalized by the pandemic, repair-cost inflation and rising total-loss frequency pressuring carrier expense ratios. The structural headwind is concentration: in US auto estimatics, one company — CCC — controls roughly 85% of the market by Tractable’s own antitrust filing (March 2024), meaning the independent AI layer must sell into, write into, and compete against the same three rails (CCC, Solera, Mitchell) everywhere it goes. And multimodal foundation models are eroding the scarcity of damage-recognition CV itself, shifting the defensible asset from models to claims-outcome data — which the estimatics incumbents hold more of.

Competitive intel

The named set and scale are in the frontmatter; the shape of the fight matters more. Tractable’s genuine edges: it is carrier-side and platform-neutral, its models were trained with cooperation from some of the world’s largest insurers across three continents, and its international position — all four top Japanese insurers, major UK/European carriers — is something no US estimatics vendor can claim. Its structural disadvantage: every distribution channel doubles as a competitor. CCC bundles Smart Estimate and Jumpstart into the workflow GEICO and 300 other carriers already pay for; Solera’s Qapter does the same on Audatex rails across Europe; Mitchell demonstrated the channel risk explicitly by adding Claim Genius alongside Tractable in 2023; Verisk owns the property system of record Tractable’s property product depends on. The 2018-2025 CCC litigation is this dynamic rendered in filings: CCC alleged Tractable used a fake appraisal firm to reverse-engineer its software; Tractable counter-alleged CCC wields an 85% estimatics share to foreclose independent AI. Both walked away in January 2025 with terms sealed — and the structural question unresolved.

History and evolution

• 2014 — Dalyac and Ranca meet at Entrepreneur First London; Tractable founded.
• 2015 — Seed (~$2M, Zetta); Adrien Cohen joins as third co-founder.
• Feb 2017 — $8M Series A (Ignition); early European deployments (Ageas, Covéa); Japan entry via Tokio Marine.
• Jul 2018 — $20M Series B (Insight).
• Sep-Oct 2018 — CCC sues Tractable in N.D. Illinois: an employee posing as “Jason Chen” of fictitious “JA Appraisal” allegedly licensed CCC ONE and disassembled it.
• Feb 2020 — $25M Series C (Georgian); $1B+ in auto claims processed.
• Oct 2020 — Admiral Seguros becomes Spain’s first insurer using AI damage assessment; MS&AD deploys across Japan.
• Jan-May 2021 — The Hartford partnership; then GEICO, announced via the WSJ.
• Jun 16, 2021 — $60M Series D at $1B; UK’s 100th unicorn; property ambitions announced.
• 2022 — Property product launches; Guidewire and Duck Creek integrations; Seventh Circuit rejects Tractable’s arbitration bid in the CCC case (June).
• 2022-23 — Layoffs and restructuring; morale dip visible on Glassdoor.
• Feb 2023 — Verisk/Xactimate property partnership; judge dismisses two counts in CCC suit, core trade-secret and fraud claims proceed.
• Jul 18, 2023 — $65M Series E led by SoftBank Vision Fund 2; valuation undisclosed.
• Mar 2024 — Tractable files antitrust counterclaims: CCC holds ~85% of estimatics and forecloses competition.
• Sep 2024 — Dalyac steps up to chairman; Venkat Sathyamurthy becomes CEO.
• Jan 23, 2025 — CCC and Tractable settle everything; claims dismissed with prejudice February 2025, terms undisclosed.
• 2025-26 — Repositioning around agentic claims AI; Everest Group Top 50 listing; NADA Show exhibitor (Feb 2026). No new funding announced.

What people say

The case for. Carriers that deployed it publicly credit real operational deltas: Admiral Seguros’ 90% appraiser-free estimates and sub-15-minute claims (2021), MS&AD citing up to two weeks cut per claim across hundreds of thousands of Japanese claims (2020), Tokio Marine building its claims journey around the AI (Insurance Innovation Reporter). The 2021 investor case — 600% growth, GEICO, first-mover CV brand — was genuine, and analysts still list Tractable among top P&C technology providers (Everest Group, 2026). Employees historically rated the ML team and mission highly, and a June 2024 Glassdoor review praises the current leadership team while acknowledging brutal SaaS market dynamics.

The complaints. Body shops are the loudest critics of photo estimating: WIRED’s April 2021 reporting (amplified by Repairer Driven News) quotes a shop owner calling 99.9% of photo estimates incorrect, the recurring complaint being that suspension, frame and hidden damage cannot be diagnosed from photos — so AI first estimates systematically understate, pushing friction into supplements. The AI Incident Database logs photo-based auto estimation producing frequently inaccurate repair prices (incident 345). The JA Appraisal episode — an employee using a false identity to access a competitor’s product — is an integrity stain the settlement does not erase from the record. Glassdoor’s critical thread from 2023 cites layoffs, chaotic reorganizations and sagging morale. And the skeptic’s structural read: the company that announced every milestone from 2017 to 2021 has announced no valuation, no revenue, no major new logo since July 2023, while its estimatics adversaries shipped competing AI to their captive bases.

Outlook: the open question

Tractable works if, over the next 18-24 months, it demonstrates that carrier-side neutrality plus superior touchless rates is a durable wedge: publicly renewed or expanded flagship accounts (GEICO, The Hartford, the Japanese majors), a disclosed revenue or funding event confirming the business grew through the silence — GetLatka’s ~$79M ARR estimate made real — and the property and auto-ecosystem lines proving the engine extends beyond a single product’s fee pool. The assets are real: logos no startup rival matches, a decade of claims-outcome training data insurers helped build, and the only meaningful international footprint in the category. It fails if the systems of record finish the enclosure — CCC and Solera’s bundled CV proving good enough for carriers already paying for the rails, Mitchell-style dual-sourcing commoditizing the AI slot, foundation models erasing the vision moat — leaving Tractable a shrinking-headcount vendor whose logical endgame is acquisition by an estimatics platform, an insurer consortium, or Verisk buying its property partner. The tells: whether a top-five US carrier signs or publicly expands with Tractable rather than CCC’s AI in 2026-27; whether a Series F (or sale) surfaces, and at what mark against 2021’s $1B; whether the Japanese majors stay exclusive; and whether the CCC détente quietly makes the independent layer a licensed feature of the incumbent it once sued.

Sources and further reading

• How we built an AI unicorn in 6 years (TechCrunch, Alex Dalyac, July 20, 2021)
• Tractable becomes world’s first computer vision unicorn for financial services (PR Newswire, June 16, 2021)
• Geico to Use Artificial Intelligence to Speed Up Car Repairs (Wall Street Journal, May 25, 2021)
• Tractable snaps up $65M led by SoftBank for car and property damage appraisals using AI (TechCrunch, July 18, 2023)
• Tractable teams up with Verisk to offer AI-powered estimates for property damage (Verisk, February 6, 2023)
• Tractable Alleges CCC Violates U.S. Antitrust Laws (PR Newswire, March 2024)
• CCC, Tractable reach settlement in five-year court case (Repairer Driven News, January 23, 2025)
• Judge grants partial dismissal of CCC intellectual property suit against Tractable (Repairer Driven News, February 15, 2023)
• WIRED features body shops’ reactions to photo estimating, photo AI (Repairer Driven News, April 14, 2021)
• From Entrepreneur First to US$1B: The Story of Tractable (Entrepreneur First, 2021)
• Q&A: Venkat Sathyamurthy, Tractable AI (Insurance Post, 2024)