Insurance / Parametric Wildfire Reinsurance & MGA · Deep dive

Kettle

The 2020 San Francisco insurtech that built a 140-million-parameter deep-learning wildfire model, launched parametric reinsurance on California brush before the Palisades and Eaton fires, replaced its own co-founder-CEO with a Root reinsurance veteran in late 2024, and now bets a ~23-person MGA can convert a modeling edge into durable capacity from PartnerRe and RLI before Munich Re, Swiss Re or a state-sponsored public model out-Kettles Kettle.

emerging

The question that decides it: Can a ~23-person modeling-first MGA convert a genuine wildfire-model edge into durable, priced reinsurance capacity through a decade in which California wildfire loss ratios are structurally repricing — before Munich Re, Swiss Re and Everest Re build equivalent deep-learning models in-house, before Cal Poly Humboldt's forthcoming public wildfire model becomes the state-sanctioned default under the Sustainable Insurance Strategy, and before a single systemic fire season (Palisades/Eaton was a warning shot) blows through the tiny book Kettle actually retains?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Kettle is a small San Francisco insurtech built entirely around one bet: that a purpose-built deep-learning wildfire model can price California brush risk accurately enough to earn reinsurance capacity from Bermuda and London, and eventually a homeowners book of its own. Founded in 2020, it raised $30M through a $25M Series A in November 2021, launched a parametric wildfire reinsurance product for California, then in 2023 partnered with wholesale broker Amwins and reinsurer PartnerRe on an excess wildfire policy for high-value California homes. In November 2024 the two founders — Andrew Engler and Nathaniel Manning — stepped aside for former Root reinsurance chief Isaac Espinoza, who has since pushed Kettle to write more of its own premium ($20M in 2024) and to expand into commercial property via a February 2026 partnership and equity investment with RLI Corp. It matters now because California’s wildfire market is being repriced in real time: State Farm, Allstate and Farmers have retreated, the FAIR Plan has ballooned to 684,388 policies and $750B of exposure by March 2026, and the Palisades and Eaton fires of January 2025 delivered $25-39B in insured losses — the exact environment Kettle was designed for, and the exact environment that could vaporize it.

Founding story

Kettle came out of two very different careers colliding at the same problem. Andrew Engler had spent roughly a decade in the traditional insurance stack — sales at Allstate, then VP of Digital at Argo Group, a specialty reinsurer where he watched underwriters price catastrophe exposure using loss triangles built for a climate that no longer exists. Nathaniel Manning had spent his career on the other side of catastrophes: as CEO of the crisis-mapping non-profit Ushahidi, as the first Chief Data Officer at USAID, and as one of the Obama White House’s inaugural Presidential Innovation Fellows working on open data for humanitarian response. Manning had spent enough time in disaster zones to see that insurance payouts, more than aid, were what actually let households rebuild — and that in wildfire-exposed California those payouts were disappearing.

The two co-founded Kettle in early 2020 in San Francisco. The pitch to True Ventures, Acrew, Homebrew, Anthemis and Inspired Capital in October 2020 was straightforward: reinsurers were retreating from California wildfire because their models could not tell brush from brush; a modern deep-learning approach trained on satellite, weather and vegetation data at fine spatial resolution could re-underwrite that risk and take the profitable slice. The seed round closed at ~$4.71M. A $25M Series A led by Acrew followed in November 2021, at the peak of the insurtech financing bubble, funded by Homebrew, True, Anthemis, Valor Equity, DCVC and Lowercarbon Capital.

How it works

Mechanically, Kettle is a machine-learning model wrapped in an MGA. Its core is a set of deep convolutional neural networks trained on approximately 130 terabytes of data drawn from roughly 40 geospatial, satellite and weather datasets — NOAA weather feeds, NASA MODIS and LIDAR imagery, vegetation and fuel-load maps, historical fire perimeters, structure characteristics, topography. The company has said a single model run touches on the order of 9 billion lines of source data and produces probabilistic wildfire-damage estimates at roughly half-square-mile resolution across California, running on a fleet of thousands of virtual machines and evaluating tens of millions of simulations against ~140 million model parameters. The output is not a single risk score but a distribution — the probability that a given parcel is inside a wildfire perimeter within a policy period, conditional on ignition and spread dynamics.

That model feeds three products. On parametric reinsurance, Kettle sells cedents (insurance companies) a contract that pays out when a wildfire crosses a defined geographic trigger, without the loss-adjustment cycle of traditional indemnity coverage. On the commercial parametric side, Kettle sells the same type of trigger directly to corporate policyholders whose properties are wildfire-exposed. On the admitted-market side, since 2023 Kettle has provided the modeling, pricing and market-matching for an Amwins/PartnerRe excess wildfire product covering California residential dwellings with replacement costs up to $20M, up to $7M of excess coverage per risk, designed to sit above the FAIR Plan or a self-insured retention.

Product and business overview

Three named product lines, all California-first. Parametric wildfire reinsurance for cedents — Kettle’s original wedge, sold through Bermuda-domiciled paper. Parametric wildfire insurance for corporates — same trigger structure, sold to enterprise property owners. Excess wildfire (Amwins/PartnerRe program) — a residential-property excess layer distributed exclusively through Amwins-appointed retail agents in California, with PartnerRe Insurance Solutions Bermuda providing capacity and Kettle providing the model, pricing and portfolio steering. On top of those, the February 2026 RLI multi-peril commercial property product — underwritten by Mt. Hawley Insurance Company (an A+ AM Best-rated RLI subsidiary) on a non-admitted basis, initially in California and Nevada, distributed through a select panel of wholesale brokers, with dedicated wildfire coverage inside a broader all-perils commercial form. Alongside the underwriting relationship, RLI took a strategic equity position in Kettle whose size has not been disclosed.

Business model and pricing

Kettle earns two kinds of revenue. As an MGA it takes ceding commissions and profit commissions on premium it writes on behalf of PartnerRe, Mt. Hawley and any other capacity provider that comes online. Historically, before the Espinoza pivot, a larger share of the revenue mix came from modeling and reinsurance-support fees rather than from insurance production. Since late 2024 the emphasis has shifted toward writing more of its own premium — CEO Isaac Espinoza told The Insurer in April 2025 that the company wrote around $20M of gross premium in 2024 and that the market is now “ripe” for its model. There is no public rate card; parametric reinsurance is priced deal-by-deal against the cedent’s exposure and Kettle’s modeled loss cost, and the excess wildfire product is priced per dwelling based on location, replacement cost and modeled parcel-level risk. The economics of the whole business rest on two variables Kettle does not control directly: the reinsurance capacity it can attract on its own paper, and the loss ratio the model actually delivers over a full wildfire cycle.

Traction over time

The shape is deliberate: a very small team, a very small book, and a strategy of stacking rated capacity partners rather than raising more equity. There has been no priced round since 2021.

Market analysis

California residential property insurance is the single most dislocated large insurance market in the United States. State Farm — the state’s largest homeowners writer — has paid $1.26 in claims for every premium dollar collected over the past nine years, disclosed cumulative California losses above $5B, and estimated ~$7.6B in direct losses from the January 2025 LA wildfires alone. The California FAIR Plan, the insurer of last resort, has grown from ~124,000 policies in 2019 to 684,388 policies and roughly $750B of exposure by March 2026, and in February 2025 was hit with a $1B assessment on admitted carriers to cover Palisades and Eaton claims. Insurance Commissioner Ricardo Lara’s Sustainable Insurance Strategy, in effect since January 1, 2025, for the first time lets carriers use forward-looking catastrophe models and pass through the net cost of reinsurance in filed rates — provided they commit to writing 85% of their statewide market share in wildfire-distressed ZIP codes.

For Kettle, this cuts both ways. The reforms legitimize the exact thing Kettle sells — a forward-looking wildfire model — and create rate-filing demand for outputs like Kettle’s. But they also invite ZestyAI, Verisk, CoreLogic and eventually the state-funded Cal Poly Humboldt public model into the same regulatory sandbox, on the same terms. And the reinsurance capacity Kettle needs is being rationed globally at the moment California most needs it.

Competitive intel

The competitive set is not one thing. Delos Insurance is the closest analog — a fellow SF MGA writing California wildfire homeowners, but focused since 2016 on the admitted-primary market rather than reinsurance, and with a larger placed book ($40B TIV) though a smaller capital raise ($16M). ZestyAI is the more dangerous competitor: it does not write policies, but its Z-FIRE model is embedded in 16 named US carriers and in the California FAIR Plan itself, and it is one of the few AI-native wildfire models cleared under the Sustainable Insurance Strategy. If the industry standardizes on Z-FIRE as the outside-the-carrier wildfire benchmark, Kettle’s core differentiation becomes negotiable. Cape Analytics ($75M raised) supplies the property-attribute layer that any competitor would combine with a peril model to reproduce much of Kettle’s underwriting stack. Behind all of them sits the coming Cal Poly Humboldt public wildfire model, and behind that, the modeling teams at Munich Re, Swiss Re and Everest Re, who ultimately decide what Kettle’s book costs.

History and evolution

• Early 2020 — Andrew Engler and Nathaniel Manning found Kettle in San Francisco.
• October 20, 2020 — Kettle announces its ~$4.71M seed led by True Ventures, with Acrew, Homebrew, Anthemis and Inspired.
• 2021 — First California wildfire parametric reinsurance products; Kettle joins the Lloyd’s Lab accelerator cohort.
• November 18, 2021 — $25M Series A led by Acrew Capital.
• April 2023 — Amwins Access launches its Excess Wildfire product with PartnerRe as capacity and Kettle as model and pricing partner; up to $7M excess on California homes up to $20M replacement cost.
• Late 2023 / 2024 — Kettle begins deliberately pivoting from selling model outputs and reinsurance support toward writing more of its own premium.
• November 5, 2024 — Kettle names Isaac Espinoza (ex-Root SVP Reinsurance) CEO; co-founders Engler and Manning transition to senior advisor roles.
• January 2025 — Palisades and Eaton fires destroy more than 16,000 structures and generate $25-39B of insured losses; California’s Sustainable Insurance Strategy takes effect the same month.
• February 13, 2026 — Kettle and RLI Corp. announce a multi-peril commercial property partnership on Mt. Hawley paper for California and Nevada, accompanied by an undisclosed strategic equity investment from RLI.
• Mid-2026 — Headcount reported around 23; no new priced financing round disclosed since 2021.

What people say

The case for. Trade press (Artemis, Reinsurance News, The Insurer, Insurance Journal, Carrier Management) has treated Kettle as a serious modeling shop rather than a marketing story, and the willingness of PartnerRe (2023) and RLI (2026) — both rated, capital-constrained specialty writers who do not lend their paper carelessly — to put balance sheet behind Kettle is the strongest external signal that the model is credible. Fast Company and CNBC profiled Kettle sympathetically during the Series A and immediate post-Series A period. Heatmap’s post-Palisades coverage treated Kettle as a real test case for whether an AI-driven wildfire insurer could function in the LA fires’ aftermath rather than dismissing it. The Sustainable Insurance Strategy’s explicit endorsement of forward-looking catastrophe models is, in effect, the state of California’s regulatory validation of Kettle’s premise.

The complaints. They are structural, not review-site. Kettle is very small: ~23 employees and ~$20M of premium written in 2024, five and a half years after founding. It has not raised a priced round since November 2021, which either reflects capital discipline or, less charitably, an unwillingness by growth investors to remark a 2021 insurtech valuation into a 2026 market that has punished the whole cohort (Root, Hippo, Lemonade all trade well below IPO). Its co-founder-CEO was replaced in November 2024 — a change the board framed constructively but which cost Kettle its most public voice and coincided with a strategy pivot. The Amwins/PartnerRe program has not been broken out publicly by premium or loss ratio, so its performance through the Palisades and Eaton fires is a black box to outsiders. And the strategic threat is real: ZestyAI is winning the model-supplier war among admitted carriers; a state-sponsored public wildfire model is coming; and the reinsurers Kettle depends on are also the entities most capable of building the same model in-house.

Outlook: the open question

The open question is whether a ~23-person modeling-first MGA can convert its wildfire-model edge into a durable, priced flow of reinsurance capacity — before larger reinsurers replicate the model, before Cal Poly Humboldt’s public model becomes the regulated default, and before a single systemic fire season blows through the small book Kettle actually retains.

For the bull case to be true, four things have to hold. First, the model has to demonstrably outperform Z-FIRE, Verisk and the incoming public model on a like-for-like basis on 2025-2026 fires — and Kettle has to publish enough of that to make brokers and regulators believe it. Second, RLI’s February 2026 equity check has to be the first of a series of rated-carrier partnerships, not a one-off, so that Kettle’s capacity stack diversifies beyond PartnerRe. Third, the Espinoza-era push into own-book premium has to compound from ~$20M toward the low-hundreds-of-millions without importing the loss ratio that made State Farm quit the state. Fourth, the Sustainable Insurance Strategy’s rate-filing rules have to reward Kettle’s granularity in a way that a public generic model cannot match.

For the bear case, the mechanism is equally concrete. Munich Re, Swiss Re and Everest Re decide Kettle is a cedent like any other and price its capacity by their own models, compressing Kettle’s underwriting margin. ZestyAI cements its position as the third-party wildfire model of record for California admitted carriers, leaving Kettle stuck in the smaller parametric-reinsurance niche it started in. The public model arrives and quietly commoditizes the “forward-looking” story. And a single bad fire season — Palisades and Eaton were $25-39B and Kettle survived on modest retained exposure, but the next one may not be — either impairs Kettle’s own book or, worse, its capacity partners’ appetite. The tells to watch: any priced financing round that resets or confirms the 2021 valuation, disclosed premium growth and loss ratios on the Amwins/PartnerRe program through a full CA fire cycle, additional rated-carrier partnerships in the RLI mold, and whether ZestyAI, Delos or a public model displaces Kettle in filed rates. Everything else — headcount, media, awards — is noise around those four dials.

How a challenger would attack it

Commoditize the model, then take the channel. Kettle’s entire edge is a proprietary wildfire model, and the file shows three routes to neutralizing it without matching it. First, ride the standards: ZestyAI’s Z-FIRE is already approved for Sustainable Insurance Strategy rate filings and embedded in ~16 carriers plus the FAIR Plan itself; a challenger licenses Z-FIRE (or waits for the free Cal Poly Humboldt public model) and gets 80% of Kettle’s analytical story at a fraction of the cost, spending its capital on distribution instead. Second, attack the capacity dependency: Kettle rents its balance sheet from PartnerRe and Mt. Hawley, has no priced round since November 2021, and runs ~23 people — a challenger arriving with committed multi-year capacity, or a reinsurer’s own paper, out-competes Kettle on the only thing brokers ultimately need: certainty the limit is there next renewal. Third, exploit the opacity: the Amwins/PartnerRe program’s loss performance through Palisades and Eaton is a black box, and in a market where State Farm ran a 126% nine-year loss ratio, “trust our model” without published results is a fragile pitch. A challenger that publishes parcel-level hit rates against 2025-2026 fire perimeters makes Kettle’s silence look like an answer.

Same playbook, new buyer

Port the parametric-trigger machine to perils and geographies with no ZestyAI. Kettle’s real invention is not wildfire per se — it is a deep-learning peril model fused to a parametric trigger and rented rated paper. California wildfire is the most contested version of that market: a state-sponsored public model coming, Z-FIRE standardizing, Delos in the admitted channel. The same architecture aimed at perils with no model-of-record — Mediterranean and Australian wildfire, US wind-driven wildland-urban interface outside California, or flood and freeze for commercial portfolios — faces none of that convergence. The nearer-term shift is buyer: Kettle sells to cedents and homeowners’ excess layers, but the FAIR Plan’s 684,388 policies and $750B of exposure make the state’s residual market itself the largest single wildfire buyer on earth, and utilities facing inverse-condemnation liability (the ignition side of the same fires) buy parametric cover at corporate scale. Kettle can’t chase these: 23 people, one state plus Nevada, capacity partners who signed up for a specific California book, and a CEO mandated to prove disciplined premium growth at home. The model generalizes; the company, as capitalized, does not.

Sources and further reading

• Deep-Learning Powered Reinsurance Company Kettle Launches — PR Newswire, October 20, 2020. Seed round, founders, initial thesis.
• Reinsurance startup Kettle raises $25m in Series A funding round — Reinsurance News, November 2021. Series A lead (Acrew), participants, product plans.
• Kettle works with Amwins & PartnerRe on California excess wildfire product — Artemis.bm, April 2023. Amwins Access program structure, PartnerRe capacity, $7M excess on $20M homes.
• Insurtechs’ Approach to Wildfire Insurance: Part II – Kettle — Insurance Journal, January 22, 2025. Model description, product suite, market context post-Palisades.
• Parametric Insurance Meets the Wildfire Peril: InsurTech Kettle Responds — Carrier Management, January 30, 2025. Product mechanics and LA fires context.
• Wildfire insurtech Kettle hires Root’s Espinoza as new CEO — The Insurer, November 2024. CEO transition, founder move to advisor roles.
• Kettle CEO Espinoza says ‘the market is ripe’ — Insurer TV, April 14, 2025. ~$20M 2024 premium; strategy pivot.
• Exclusive: Kettle, RLI forge commercial property push with wildfire focus, equity backing and E&S expansion — The Insurer, February 13, 2026. RLI/Mt. Hawley partnership and strategic equity investment.
• California’s Sustainable Insurance Strategy: What it means for homeowners — United Policyholders, 2025. Reforms in effect January 1, 2025 including forward-looking cat models.
• Industry insured losses for Los Angeles wildfires — Milliman, 2025. Palisades/Eaton insured loss estimates.
• An Insurance Startup Faces a Major Test in Los Angeles — Heatmap News, January 2025. Kettle’s exposure to and framing around the LA fires.
• Our Kettle company profile — Tracxn, 2026; and Revelio Labs employee count (~23, July 2026).