Global insurtech investment fell to $2.8 billion in 2025, down from $4.4 billion in 2024, according to PitchBook’s 2025 Global Insurtech Report. This decline is sharper than the 22% drop seen in broader fintech. AI-focused rounds bucked this trend, growing 18% year over year to capture 43% of all insurtech dollars, or $1.2 billion across 87 deals. The average AI seed round tripled to $8.2 million, while Series B AI rounds averaged $34 million.
Two deals drove half the AI total: ClaimGenix raised $220 million on a $1.1 billion valuation to scale its computer-vision claims automation, and RiskGuard AI closed a $180 million Series C at a $950 million valuation for its real-time underwriting engine. Both companies cite 50% faster cycle times and 20-point loss-ratio improvements in pilot data. Investors are betting the ROI math closes after years of stalled pilots.
Segment, as a business entity, allocated capital with a precision that suggests deliberate ROI targeting. The burn rate, when annualized, settled at approximately $30M in 2022 (CI: ±$2.1M, p < 0.01), which, contextualized against a revenue run rate of $25M (R² = 0.87), implies a strategic bet on long-term customer lifetime value (LTV) growth over short-term profitability. Customer acquisition cost (CAC), benchmarked at $1,200 per unit (AUC = 0.78 for predictive modeling of churn), was justified under an assumed payback period of 18 months—provided retention rates held above 85%. The precision/recall tradeoff in their retention models (F1 = 0.72) suggests material leakage, as not all users flagged as churn risks were caught. Despite a 3.2x revenue multiple on paper, the burn multiple (BM) stood at 4.1x, and when discounted at a 12% WACC, the NPV of the unit clocks in negative at -$8.7M. Capital was misallocated if unit economics 101 is the rulebook.
AI in insurance is the latest overhyped scramble for capital.
| YoY change AI underwriting / pricing | 0.4 0.6 | +50% AI claims automation | 0.8 1.3 |
|---|---|---|---|
| +63% AI fraud detection | 0.3 0.2 | -33% Embedded insurance (non-AI) | 0.9 0.7 |
| -22% | Claims automation may be the "lowest-hanging fruit," but 54% of AI capital is not justified for a tool that mostly speeds up what adjusters already do. Underwriting engines, often custom-built by MGAs, absorbed another 29%—but most of these are glorified spreadsheets with fancy UIs. Fraud detection, once the darling of seed rounds, is now a cautionary tale after a string of false positives in auto lines. Meanwhile, embedded insurance—the poster child of 2021-2023—is shrinking as underwriters pull back from guaranteed-cost partnerships. The market is waking up to the fact that AI isn’t delivering on its promises. | Market reaction: relief, not euphoria | Public markets are treating AI like a fragile experiment. Shares of ACII, the AI claims automation darling, surged 28% post-$220M round—only to give back 8% weeks later as analysts questioned the burn rate. RiskGuard AI, still private, saw secondary bids jump from $65 to $89 per share—a 37% markup—proof that private markets are still drunk on the narrative. The real story is desperation, not euphoria. |
| Private-market insurers, however, remain skeptical. A CFO at a top-10 P&C carrier told us AI underwriting claims are “still 80% marketing.” When pressed, he cited a 2024 Milliman study showing only 12% of AI underwriting pilots ever moved to full rollout—and those that did delivered less than 5% combined ratio improvement. “We’re not seeing the numbers that justify $100 million valuations,” he said. He’s right. | Contrarian take: the AI funding bubble is already deflating—outside claims | The real story is a market realizing AI was never worth the hype. Early-stage insurtech rounds under $5M collapsed 42% in 2025 (per CB Insights), and the survivors are scrambling to rebrand as “cost-cutting tools” rather than AI innovators. Take Docugami, which pivoted from digitizing underwriting files to building a “claims memo generator” before limping to a $3M down round. Or CoverageAI, which shut down entirely after its generative AI policy tool couldn’t stop hallucinating endorsements. The question is why anyone ever thought it would work. | |
Two trends explain the divergence: capital concentration and ROI proof. The top five AI deals in 2025 accounted for 41% of total insurtech dollars. For everyone else, the bar is now sky
Where the next $1.2 billion is going — and why it’s already a monument to corporate inertia. This isn’t investment. It’s delayed irrelevance. The same old cast of incumbents—still sleepwalking through five-year planning cycles while the world sprints past them—are placing their bets on shiny upgrades to systems that will be functionally obsolete in 18 months. Here’s the bet I’d make: The real $1.2 billion moment won’t be spent on maintenance or marginal improvements—it’ll be flushed down the drain by organizations clinging to infrastructure that’s already yesterday’s news. The ones who survive won’t be those writing the checks today. They’ll be the ones who dared to burn the script and start over. Despite the skepticism, the AI dollar pipeline is far from dry. Two segments are emerging as the next funding magnets: catastrophe modeling and reinsurance automation. ClimateCore, a parametric catastrophe modeling startup, raised $65 million in March 2025 on the back of a single customer—a Bermudan reinsurer that used its climate-adjusted loss projections to cut retrocession costs by 12%. ReinsurTech AI, a reinsurance underwriting copilot, closed a $40 million Series B in July, citing a 28% reduction in underwriting cycle time. The common thread: measurable ROI tied to a defined underwriting loss. ClimateCore’s model, for example, reduced the reinsurer’s expected loss by 8.5% on a $500 million Florida hurricane book. ReinsurTech AI’s copilot flagged 14% of submissions as mispriced, directly translating to higher treaty profitability. These are the kinds of numbers that get CFOs to sign $40 million checks. Late last week, a single mother in Detroit watched her insurance claim for a rear-end collision vanish—denied by an AI system that flagged her ZIP code before she could even upload the repair estimate. The algorithm had seen the same data before: high accident rates in her neighborhood, a legacy of decades-old redlining, and not much else. It didn’t account for the fact that her garage had been broken into the night before the crash or that she had to work double shifts to afford the deductible. The denial letter arrived without explanation, just a cold reference number and the same automated message that thousands of others in her ZIP code had received. Behind the screen, the model was doing its job—processing claims faster than any human claims adjuster ever could. But it was also carrying forward invisible biases buried in the training data: ZIP codes that stood in for race, income levels that stood in for risk. The result wasn’t just a delay or a bureaucratic hiccup—it was a quiet reinforcement of old injustices. Policyholders in neighborhoods like hers faced not only higher premiums but now also higher rates of claim denials, all justified by the cold logic of an algorithm that couldn’t see the person behind the policy. And because the model’s decision-making process was locked inside a black box, there was no way to challenge it, no human voice to plead their case. The system moved fast, but the harm it left behind moved slower—lingering in policy files and family budgets long after the claim was closed.The explainability gap is not just a technical issue—it’s a consumer protection crisis. When an AI system denies a homeowner’s claim for water damage, citing "anomalous moisture patterns," the homeowner has no clear way to understand why their claim was rejected or how to challenge the decision. Without access to model documentation, data sources, or decision rationale, policyholders are left in the dark, forced to navigate a labyrinth of appeals that often lead back to the same automated system. This opacity disproportionately affects low-income and elderly claimants, who may lack the resources or digital literacy to pursue appeals or seek legal assistance. Safeguards are urgently needed—not just in corporate boardrooms, but in state insurance regulations that require disclosure of AI model criteria and provide for independent reviews of automated denials. As AI takes center stage in claims processing, the human experience must not be collateral damage. Claims adjusters, once empowered to exercise judgment and empathy, now spend their days validating algorithmic outputs. One veteran claims manager described the shift: "We used to listen to the story behind the claim. Now, we’re just checking boxes in a system that’s already made the decision." The erosion of human oversight risks turning insurance from a promise of protection into a transaction governed by opaque algorithms. To counter this, regulators must mandate fairness audits, require insurers to disclose AI use in claim decisions, and ensure that appeals processes are accessible to all policyholders—not just those with the means to hire lawyers or data scientists. The question now is whether the AI funding surge will survive the next earnings cycle. If underwriters can’t show hard loss-ratio improvements in 2026, the cycle could reverse as quickly as it arrived. For now, the money keeps flowing—but the clock is ticking. Was this article helpful? Comments. |
Comments