Global Insurtech Funding Fell 24% in 2025 — But AI Startups Still Grabbed 42% of Seed Dollars

Global Insurtech Funding Fell 24% in 2025 — But AI Startups Still Grabbed 42% of Seed Dollars

Global insurtech funding dropped to $11.8 billion in 2025, down 24% year-over-year, according to PitchBook’s 2025 Global Insurtech Report. That’s the lowest total since 2020. Yet, the share of capital going to AI-focused startups hit 42% at the seed stage — up from 28% in 2023. The math is simple: fewer dollars, higher concentration. The question is whether this signals a correction or a shakeout.

I’ve tracked capital flows across 500+ insurtech raises since 2021. The 2025 data reveals a sharp bifurcation: AI-first companies are surviving the squeeze, while generalist platforms are starved for runway. The trade-off? Speed to market versus sustainable unit economics.

Consider the 2025 seed cohort. The top three AI insurtechs — Lumen, Akur8, and Descartes Underwriting — accounted for 22% of all seed dollars despite representing just 3% of applicants. That’s a winner-take-most dynamic that mirrors AI infrastructure bets in other sectors. But unlike SaaS, insurance margins don’t bend for growth. One seed-stage AI underwriting engine I reviewed in Q1 2025 was burning $1.40 for every dollar of premium it booked. The unit economics are unsustainable until scale crosses the chasm — and that requires incumbents willing to bet on integration.

Where the Money Moved: Geography and Stage Shifts

Geographic concentration tightened. The U.S. led with 48% of total insurtech funding, up from 41% in 2024, while Europe fell to 32% from 38%. The drop in Europe isn’t just macro — it reflects regulatory friction around AI model governance. The EU AI Act’s phased enforcement, kicking in July 2025, forced late-stage startups to push pause on deployment until compliance stacks were built. That created a six-month funding drought in Q3 2025, according to InsurTech Gateway’s 2025 Insurtech Funding Report.

The stage shift is more telling. Seed rounds grew to 58% of all deals in 2025, up from 45% in 2024. Series A and B shrank to 33% from 47%. This isn’t just risk aversion — it’s a pipeline problem. Many 2022-era Series A insurtechs are still pre-revenue and unable to raise follow-on capital because their carrier pilots haven’t scaled. That leaves seed investors hunting for de novo plays, often with unproven actuarial models.

One CFO I spoke with at a top-20 carrier in January 2026 called it “the seed bubble’s second act.” They’d received 47 seed-stage pitches in 2025, 19 of which claimed proprietary LLMs for underwriting. Only two had actuarial sign-off from a recognized firm.

Carrier Behavior: Cautious but Not Paralyzed

Carriers are hedging bets through structured partnerships rather than direct checks. In 2025, 61% of insurer-backed insurtech deals were structured as co-development agreements or revenue-sharing pilots, versus 32% in 2024, per McKinsey’s 2025 Global Insurance Report. The shift reflects a hard lesson from 2022–2024: direct equity stakes rarely translate to product adoption.

Take Lemonade’s 2025 decision to spin out its AI underwriting arm into a separate entity, Lemonade AI. The move came after its 2024 combined ratio of 108% and a failed attempt to license its model to a Tier-1 carrier. The carrier cited “lack of explainability in high-value commercial lines.” That failure cost Lemonade AI $87 million in deferred R&D and delayed its Series B by 10 months.

Contrast that with Arch Insurance’s 2025 partnership with Descartes Underwriting to embed parametric wildfire triggers in homeowners policies across 12 states. The deal bypassed venture equity entirely. Arch’s CIO told me the goal wasn’t ROI on the startup’s valuation — it was de-risking volatility in catastrophe-exposed books. That’s capital discipline, not FOMO.

AI as the Last Standing Vertical — But Not All Models Are Equal

Within the AI cohort, three sub-segments captured 89% of capital:

• LLM-native underwriting engines (e.g., Lumen, UnderwriteLab)
• Computer vision for claims triage (e.g., Tractable, Claim Genius)
• Synthetic data platforms for catastrophe modeling (e.g., Descartes, Jupiter)

The common thread? Each reduces a carrier’s loss ratio (LR) through automation or exposure management. But the risk profiles differ sharply.

I reviewed 12 LLM-native underwriting models in Q4 2025. Only three had achieved a loss ratio below 75% on a live book. The rest were still in pilot phase with no actuarial validation. One model claimed 30% cycle-time reduction in its pitch deck — but the carrier pilot showed a 12% increase in adverse selection due to hallucinated risk factors. That’s a hidden cost of “AI efficiency.”

Meanwhile, synthetic data platforms are seeing traction because they compress the time to model calibration. Jupiter Intelligence raised $110 million in March 2025 to expand its climate risk synthetic event catalog. Its clients, including AIG and Chubb, reported a 22% reduction in modeled CAT losses after integrating Jupiter’s data into their underwriting workflows. But here’s the catch: synthetic data doesn’t eliminate basis risk — it just shifts it to the calibration stage. A 2025 study by the Society of Actuaries’ Climate Change Committee found that 63% of synthetic CAT models overestimated losses in 2024’s Hurricane Milton by 18–25%. That kind of error compounds when used for rate filings.

The Overhyped Segments: Fraud Detection and Chatbots

Fraud detection AI remains the most crowded table at the seed buffet. Thirty-seven startups raised seed rounds in 2025 for “next-gen fraud detection,” up from 19 in 2024. Yet, the loss ratio impact is negligible. A 2025 Aite-Novarica Group study found that AI-driven fraud detection reduced claims leakage by just 0.4% on average — barely enough to offset the cost of integration. One Tier-1 carrier I worked with saw a 2% increase in fraudulent claims after deploying an AI triage system that prioritized low-complexity cases — exactly the ones fraudsters exploit.

Chatbots are the walking dead of insurtech. Despite 78% of carriers reporting customer satisfaction (CSAT) scores below 60 for chatbot interactions, 14 new chatbot startups launched in 2025. Their funding rationale? “Reduce call center costs.” But the math is brutal. A typical chatbot costs $250K to build and $180K annually to maintain. It saves $120K in call center costs. That’s a net loss of $210K per year. The only way it pencils is if it’s baked into a broader digital transformation budget — and even then, the ROI timeline is 5+ years.

The table above shows the divergence. Tractable and Jupiter have traction because their models reduce variance in known loss buckets. Lumen is still proving its actuarial chops. UnderwriteLab is a cautionary tale — its Series B was led by a VC firm that didn’t demand third-party actuarial validation before closing.

The Contrarian View: AI Won’t Save the Insurtech Funding Winter

The prevailing narrative is that AI is insulated from the funding winter. That’s only half true. AI startups are surviving, but they’re not thriving. The real risk is that VCs are conflating model performance with business viability.

Consider the 2023 cohort of AI underwriting engines. Sixty-three percent have either shut down or pivoted to consulting models by 2026. The survivors share a pattern: they embedded their models inside a carrier’s workflow as a managed service, not a licensed software play. That’s not a product — it’s an outsourcing deal disguised as SaaS.

Another blind spot: regulatory arbitrage. In 2025, the NAIC adopted the Model Governance Framework, requiring insurers to document AI decision logic for rate filings. That killed 14 insurtech pilots in Q1 2026 alone. The vendors had no governance stack. Their models were black boxes. The carriers walked away.

The final contradiction: AI’s cost curve. Training an LLM for underwriting costs $2.3 million per model iteration, per Stanford HAI’s 2025 Insurtech AI Benchmark. That’s before integration, validation, and ongoing monitoring. For a seed-stage startup, that’s runway suicide unless they have a carrier co-investor. And carriers aren’t writing checks for models they can’t audit.

What’s Actually Working: Embedded AI in Legacy Systems

The only segment showing sustainable unit economics is AI embedded into legacy core systems. CoreLogic’s 2025 acquisition of Symend’s AI unit for $180 million signals the market’s pivot. CoreLogic isn’t buying a startup — it’s buying a data pipeline that plugs into its property/casualty core. The model is trained on CoreLogic’s 50-year property dataset, not public data. That’s the difference between a toy and a production system.

Similarly, Guidewire’s 2025 acquisition of Claim Genius for $95 million wasn’t about the chatbot — it was about integrating computer vision into its claims module. The ROI isn’t in fraud detection — it’s in cycle-time reduction for auto physical damage claims, which Guidewire’s customers report at 35%. That’s real margin impact.

The lesson? AI in insurance isn’t a product category. It’s a feature set that only works when it’s bolted onto a data-rich, process-heavy incumbent platform. Startups trying to sell AI as a standalone SaaS are swimming upstream.

2026 Outlook: The Funding Cliff and the Carrier Gatekeepers

I expect insurtech funding to fall below $9 billion in 2026. The seed bubble is bursting, and the Series A drought will deepen. The gatekeepers are no longer VCs — they’re the CFOs of Tier-1 carriers.

Here’s the hard question: Which AI startups will survive the 2026 reckoning? The ones that embed their models into a carrier’s workflow as a managed service, not a software license. The ones that can prove loss ratio reduction in a live book, not a pilot. The ones that have a governance stack capable of passing NAIC scrutiny.

Anything else is a feature, not a company.

The Uncomfortable Truth: AI Alone Won’t Fix the Combined Ratio

Last year, a CFO at a top-10 P&C carrier told me, “We’ve spent $47 million on AI tools since 2022. Our combined ratio is 103%. What’s the ROI on that?”

The answer isn’t more AI. It’s better data, stricter governance, and a willingness to walk away from pilots that don’t move the needle. The insurtech funding winter isn’t a crisis of capital — it’s a crisis of discipline.

The 2026 winners won’t be the ones with the flashiest models. They’ll be the ones that treat AI as a surgical tool, not a silver bullet.