Teardown

Insurance · Deep dive

Shepherd

AI-native managing general underwriter for commercial construction and the AI-infrastructure buildout — casualty, excess and builder's risk coverage priced off live jobsite telemetry from Procore, Autodesk, OpenSpace and Samsara, underwriting the data centers, chip fabs and energy projects behind the AI boom on other carriers' paper.

emerging

The question that decides it: Shepherd books commission on other carriers' paper — Core Specialty's StarStone since 2022, now Intact as anchor investor and long-term capacity provider — and its 7x revenue growth (24 months to Mar 2026) rides the AI data-center construction cycle, the single most crowded segment in commercial insurance by its own CEO's admission. Does underwriting off live jobsite telemetry (Procore, OpenSpace, Samsara feeds priced into casualty terms) produce loss ratios visibly better than conventional construction books before the soft market and excess-casualty severity force its capacity partners to reprice — turning Shepherd into the program Intact scales rather than the front end Intact absorbs once the workflow is proven?

My take

HQ
San Francisco, CA
Founded
2020
Ownership
Private, venture-backed (Y Combinator W21)
Funding
$67M total — $6.15M seed led by Spark Capital (Sep 2021); $13.5M Series A led by Costanoa Ventures (Feb 2024); $42M Series B led by Intact Private Capital (Mar 2026). Announced rounds sum to ~$61.7M; the remainder is YC and unannounced top-ups
Valuation
Undisclosed at every round
Revenue
Undisclosed. Company reported GWP grew 5x year-over-year in the ~18 months to Feb 2024, and revenue grew more than 7x in the 24 months to Mar 2026; no absolute GWP, revenue or loss-ratio figure has ever been published
Headcount
~75 across five offices (company, Mar 2026); actively hiring production underwriters nationwide
Screen
Fast riser — founded 2020, $67M raised, US-based
Published
2026-07-23
Web
www.shepherdinsurance.com
Elsewhere
LinkedIn · Crunchbase

Founders and leadership

  • Justin Levine Co-founder & CEO

    Georgia Tech civil engineering; worked in construction risk before co-founding TradeTapp, a subcontractor risk-qualification SaaS acquired by BuildingConnected, which Autodesk then bought in 2018. At Autodesk he ran risk management strategy and watched contractors generate safety and quality data that insurers never priced. Shepherd is the second act built directly on that arbitrage.

  • Mohamed (Mo) El Mahallawy Co-founder & CTO

    Former Airbnb product engineer. Met Levine through the On Deck Founder Fellowship during COVID — Levine cold-messaged him on Slack; Mahallawy took the first Zoom from lockdown in Toronto — and they spent months validating the idea before committing. Owns the AI submission-ingestion and pricing platform.

  • Stephen Buonpane Co-founder & Chief Insurance Officer

    Roughly three decades at Chubb, latterly EVP and construction industry practice leader. The insurance adult in the room: brought carrier credibility, broker relationships and the underwriting discipline capacity providers require before lending their paper to a startup MGU.

Snapshot

Shepherd is a San Francisco managing general underwriter that sells commercial property and casualty insurance — primary casualty, excess casualty, builder’s risk — for construction and, increasingly, the AI-infrastructure buildout: data centers, semiconductor fabs, utility-scale solar and battery storage. It underwrites on other carriers’ paper, wins deals on speed — indications in minutes to hours against an industry norm of weeks — and prices off live jobsite data from Procore, Autodesk, OpenSpace, DroneDeploy and Samsara. In March 2026 it raised a $42M Series B led by Intact Private Capital — the venture arm of one of the world’s largest commercial insurers, which also becomes a long-term capacity provider — bringing total funding to $67M, alongside disclosed revenue up 7x in 24 months, 1,500+ policies, 600+ clients and $400B+ in insured project value. It is the cleanest live test of a hypothesis insurers have circled for a decade: that construction software telemetry can price risk better than actuarial tables built on stale loss runs.

Founding story

Justin Levine came up through construction risk, not insurance. A Georgia Tech civil engineer, he co-founded TradeTapp, a SaaS tool for vetting subcontractor risk; BuildingConnected acquired it, Autodesk acquired BuildingConnected in 2018, and Levine ended up running risk management strategy inside Autodesk’s construction division. There he watched the software wave — Procore, OpenSpace, drones, wearables — demonstrably make jobsites safer while insurance quoting still ran on emailed PDFs and stale loss runs. The gap between what the data showed and what underwriters priced became the company.

He met Mohamed (Mo) El Mahallawy, a former Airbnb product engineer, through the On Deck Founder Fellowship in 2020 — a cold Slack message, then a Zoom with Mahallawy locked down in Toronto. Stephen Buonpane, Chubb’s construction practice leader after roughly three decades there, joined as third co-founder and chief insurance officer — the hire that made carriers take a two-person startup seriously. Shepherd went through Y Combinator’s W21 batch but deliberately delayed demo day; licensing an insurance entity and winning capacity is a long road. The $6.15M seed led by Spark Capital arrived in September 2021, with Procore itself on the cap table.

How it works

Shepherd is an MGA/MGU: it holds underwriting authority delegated by risk-bearing carriers, prices and binds policies within agreed parameters, and earns commission — the carrier and its reinsurers keep the premium and the losses. Its first capacity deal (February 2022) was with Core Specialty: excess liability on StarStone Specialty paper, up to $10M in limits, most policies attaching within the first $25M of a contractor’s tower, backed by a reinsurance panel including Greenlight Re, also an investor. The Series B adds Intact as anchor investor and long-term capacity provider — significant because capacity, not demand, is the binding constraint on any MGA.

The pitch is speed and data. A broker at Marsh, Aon, Lockton or Alliant emails a submission — loss runs, exposure schedules, project lists, in whatever messy format. Shepherd’s AI parses the files into structured datasets and produces a priced indication before an underwriter touches the account; the company claims minutes where traditional carriers take weeks, and advertised 24-hour submission-to-indication turnaround by February 2024. The second mechanism is telemetry: through Shepherd Savings, insureds connect their Procore, Autodesk, OpenSpace or Samsara accounts, and Shepherd underwrites off incident tracking, inspection rates and fleet data as the project actually runs, offering up to 25% premium credits upfront for verified tech adoption. Levine’s stated roadmap (March 2026) is an autonomy ladder: from an underwriter working ~20 accounts a month to orchestrating 200, with fully agentic submission intake — email in, price out, human review only at the last mile.

Product and business overview

Shepherd launched single-product — excess casualty, February 2022 — and by the Series A (February 2024) had four products spanning all three primary casualty lines: general liability, commercial auto, plus excess. Builder’s risk (course-of-construction property) launched April 2025, opening the property side of data-center construction; a renewable energy and power casualty vertical is live as of March 2026, with workers’ comp expanding and renewables property announced as next. The Series A also launched Shepherd Compliance, AI software for contractor insurance-compliance workflows — an echo of the TradeTapp playbook, embedding software beyond the policy. Distribution is broker-led: the company said in 2026 it supports virtually all top-20 national retail construction brokers across nearly 200 offices. Customers skew large — general contractors, specialty builders, energy developers, and, per the Series B release, frontier AI labs, chip manufacturers and hyperscalers.

Business model and pricing

Revenue is MGA economics: commission on gross written premium — typically low-teens to high-twenties percent across the MGA industry depending on line and services, sometimes with profit-share contingent on loss performance. Shepherd publishes neither its commission rates nor premium totals, and carries little or no underwriting risk on its own balance sheet; Core Specialty, Intact and the reinsurers bear the losses. Policy pricing is bespoke, but the market context is real money: brokers put builder’s risk premiums for a single $2B AI data center at $3M-$10M a year (Hotaling Insurance Services, 2026). Shepherd’s advertised lever is the up-to-25% premium credit for telemetry-verified insureds — a bet that the data lets it discount aggressively for good risks and still hand capacity partners a better loss ratio than the market. The bet is unproven in public: no loss ratio has ever been disclosed, though Levine says Intact examined loss ratios before leading the round.

Traction over time

DateMarkerDetail
Jan 2021Y Combinator W21Demo day deliberately delayed; insurance licensing underway
Sep 2021Seed$6.15M led by Spark Capital; Procore, Susa, Greenlight Re participate
Feb 2022First capacity + productCore Specialty deal; excess casualty on StarStone paper, up to $10M limits
Feb 2024Series A$13.5M led by Costanoa; GWP up 5x YoY; one product grown to four; Shepherd Compliance launches
Mar 2024Scale-up markersMoves out of WeWork into own SF office; partnerships with Marsh, Aon, Lockton, Alliant cited
Apr 2025Builder’s risk launchEnters course-of-construction property; renewables vertical follows
Mar 2026Series B$42M led by Intact Private Capital; revenue up 7x in 24 months; 1,500+ policies, $400B+ insured project value, 600+ clients; ~75 staff, five offices

Read the numbers carefully: $400B is insured project value, not premium; Shepherd has never published GWP, revenue or a loss ratio in absolute terms, and the 7x compounds off an undisclosed base. The credible external signal is structural — a major global commercial insurer led the round after seeing the book, and the same investors kept re-upping across three rounds.

Market analysis

Construction insurance globally was roughly $39B in gross premium in 2024, projected to ~$92B by 2034 at an 8.9% CAGR (Market.us, 2024), North America about 37% of it; IBISWorld put US commercial construction insurance revenue at ~$50.3B in 2024. The structural driver Shepherd bet the company on is the AI-infrastructure capex cycle — the largest US construction cycle in a generation, in Levine’s framing — and every data center, fab and power project in it needs builder’s risk and casualty towers, built faster than legacy underwriting moves. The countervailing forces: commercial construction casualty is softening for well-performing accounts (broker updates, 2025-26), eroding pricing power exactly where Shepherd competes; excess-casualty severity keeps making carriers withdraw or shrink deployed limits (WTW, Amwins 2025-26), threatening the capacity MGAs depend on; and a widely noted surplus of new MGAs means Shepherd’s structure, if not its data, is easily replicated.

Competitive intel

The incumbents — Zurich, Travelers, AXA XL, Chubb — dominate large-account construction P&C with decades of contractor loss data and balance sheets that hold risk net; all of them are chasing data-center premium, which is why Levine calls it one of the most competitive segments in commercial insurance. Procore Risk Advisors is the most conceptually threatening rival: Procore — a Shepherd seed investor and data partner — runs its own construction insurance brokerage (2022) with exclusive Allianz- and Swiss Re-backed programs priced off Procore platform data: the same telemetry thesis, operated by the company that owns the telemetry. Foresight (~$59M raised per Tracxn; QBE capacity) runs the analogous play in construction workers’ comp, the line Shepherd only began entering in 2025-26. Ledgebrook and a wave of post-2022 E&S MGAs sell AI-assisted speed-to-quote as a horizontal, compressing its shelf life as a differentiator. On hyperscale data-center property, FM Global and Munich Re have done engineering-led underwriting of mission-critical facilities for decades — functionally the incumbent version of Shepherd’s data pitch. Shepherd’s edge is the combination — construction-native founders, structured telemetry integrations, and now Intact’s paper — rather than any single element, none of which is individually unique.

History and evolution

Notably absent: no announced pivots, capacity losses or layoffs — but also near-zero independent scrutiny for a company touching $400B of project value.

What people say

The case for. Broker behavior is the strongest evidence: by the company’s account (2024-26), virtually all top-20 retail construction brokers — Marsh, Aon, Lockton, Alliant among them — submit business, and its customer stories tout submission-to-indication speed as the reason accounts move. Investors who can see the actual book keep doubling down: Spark, Costanoa, Greenlight Re and Intact all re-upped across rounds, and Intact — a carrier that has watched a decade of insurtech fail — put its own paper behind the platform after reviewing loss ratios. BuiltWorlds’ 2024 profile cast Shepherd as the credible version of the tech-priced-insurance dream because it is an actual insurance operation, not a lead-gen layer.

The complaints. Independent review surface is close to zero — and polluted: the “Shepherd Insurance” pages on Glassdoor and elsewhere (4.0 rating, 26 reviews, a 2024 Besnard merger, ~468 employees) belong to an unrelated Indiana agency, shepherdins.com, a genuine diligence trap. For the actual company there are no G2/Capterra reviews and no meaningful Glassdoor record at ~75 employees; every operating metric — 7x revenue, 5x GWP, $400B insured value — is company-reported, off undisclosed bases, with no published loss ratio. The categorical criticisms bite hardest: broker commentary through 2025-26 (WTW, Amwins, IMA) describes an over-supplied MGA landscape where capacity is being withdrawn from excess casualty as severity worsens — the precise line Shepherd started in — and softening rates on good commercial construction accounts, which squeezes commission income that scales with premium. And delegated authority has its oldest pattern: when an MGA’s book performs, the capacity provider is tempted to take it in-house; when it doesn’t, capacity vanishes. Intact’s dual role as lead investor and capacity provider hedges the second outcome while embodying the first.

Outlook: the open question

For Shepherd to work, telemetry-priced underwriting must show up in the loss ratio — concretely: Intact and Core Specialty must see, over the 2026-2028 casualty development window, that Shepherd’s book outperforms conventional construction books by enough to justify expanding delegated capacity and tolerating its 25% good-risk discounts, before soft-market pricing and incumbent competition for AI-infrastructure accounts compress the commission pool it lives on. The bull case is strong: a founding team with the rare tri-skill combination (contractor risk, carrier underwriting, consumer-grade engineering), integrations with the construction-software oligopoly that would take a carrier years to replicate, distribution through every major broker, an anchor carrier that funded it after seeing the actuals, and a once-in-a-generation construction cycle concentrated in exactly the complex, fast-moving projects where slow underwriting loses. If the loss-ratio evidence lands, Shepherd becomes the underwriting layer for the physical AI economy, adding lines — workers’ comp, renewables property — until a broker never needs a second market.

What would sink it: the data proving decorative. Casualty is long-tail; claims from policies written in 2023-24 will still be developing in 2028, and if Shepherd’s telemetry-discounted book develops like everyone else’s, it is a fast quoting shop earning thinner commissions in a softening market — capacity reprices, the 25% credits become adverse selection, and the incumbents’ data-center appetite finishes the job. The tells: whether Intact expands or merely maintains capacity through 2027; whether Shepherd ever publishes a loss ratio or premium figure; whether workers’ comp and renewables property launch on schedule; whether Procore deepens the partnership or routes its data advantage exclusively through Procore Risk Advisors; and whether the AI-infrastructure capex cycle — the tide lifting every number on this page — holds. Shepherd raised into strength in March 2026; the book it wrote during the boom will render the verdict.

How a challenger would attack it

The wedge is the data source, not the underwriting. Shepherd’s edge rests on telemetry it doesn’t own — Procore, Autodesk, OpenSpace, Samsara feeds — and Procore has already shown it will compete for the same position through Procore Risk Advisors, with Allianz and Swiss Re paper. A challenger would go where Shepherd is structurally exposed: cut an exclusive or preferential data deal with one of the platform owners, or embed underwriting directly inside the construction software rather than asking insureds to connect accounts. Second vector: balance sheet. Shepherd clips commission on other carriers’ paper, so a challenger operating as a full-stack carrier — or an incumbent like Chubb standing up a telemetry program — can hold risk net, undercut on price through the soft market, and outlast an MGA whose capacity can be repriced or withdrawn in a severity cycle. Third: speed is already commoditizing — Ledgebrook and the post-2022 E&S wave sell AI-assisted quoting as a horizontal — so a challenger wouldn’t fight on turnaround time; it would fight on proof, publishing loss ratios Shepherd has never disclosed and forcing the question its capacity partners will eventually ask anyway.

Same playbook, new buyer

Telemetry-priced insurance travels to any industry with a software oligopoly and stale actuarial tables. The nearest adjacencies: workers’ comp for safety-instrumented industries beyond construction (Foresight is already there with agriculture), operational insurance for the data centers Shepherd only covers during construction — the handoff from builder’s risk to FM Global at commissioning is a seam a new entrant could own end-to-end — and mid-market contractors, since Shepherd’s book skews toward large GCs, hyperscalers and frontier AI labs while smaller builders running the same Procore stacks go unpriced. Geography is the other opening: the playbook is US-only today, and European or Gulf infrastructure buildouts run on the same Autodesk and Procore telemetry with none of the delegated-authority competition. Shepherd won’t follow easily because its capacity relationships (Core Specialty, Intact) are line- and market-specific, its ~75 people are consumed by the AI-infrastructure land grab, and every expansion requires a new capacity negotiation — the binding constraint its own model names.

Sources and further reading

Capital history

DateRoundAmountValuationLead(s)
Jan 2021 Pre-seed (Y Combinator W21) Standard YC terms Undisclosed Y Combinator (Shepherd deliberately delayed demo day)
Sep 2021 Seed $6.15M Undisclosed Spark Capital (Natalie Sandman); Susa Ventures, Procore Technologies, Y Combinator, Greenlight Re, Oldslip, fintech angels
Feb 2024 Series A $13.5M Undisclosed Costanoa Ventures (lead); Intact Ventures, Era Ventures, Greenlight Re, Spark Capital
Mar 2026 Series B $42M Undisclosed Intact Private Capital (lead); Spark Capital, Costanoa Ventures, others

Investors / owners: Intact Private Capital, Spark Capital, Costanoa Ventures, Intact Ventures, Era Ventures, Greenlight Re, Susa Ventures, Procore Technologies, Y Combinator, Oldslip

Competitive set

  • Zurich, Travelers, AXA XL, Chubb (incumbent construction P&C) — The incumbents own the large-account construction casualty and builder's risk market Shepherd is wedging into, with balance sheets in the tens of billions and decades of contractor loss data. Levine himself concedes AI infrastructure is one of the most competitive segments in commercial insurance — every one of these carriers wants the same data-center premium, and they can hold risk net rather than clip commission.
  • Procore Risk Advisors — Procore — a Shepherd seed investor and Casualty Pro partner — launched its own construction insurance brokerage in 2022, with exclusive programs backed by Allianz and Swiss Re that price off Procore platform data. The same thesis as Shepherd Savings, run by the company that owns the data source. Partner today, structural rival for the telemetry-underwriting position.
  • Foresight Risk & Insurance Services — Tech-enabled workers' comp MGA for construction, agriculture and other safety-critical industries; ~$59M raised (Tracxn, 2026), capacity from QBE North America, built around embedded Safesite safety software. Attacks the adjacent line Shepherd only began expanding into in 2025-26.
  • Ledgebrook — Boston E&S MGA founded 2022, generalist rather than construction-specialist, selling speed-to-quote to wholesale brokers off a modern stack. Evidence that fast AI-assisted quoting alone is becoming table stakes across the MGA landscape rather than a durable edge.
  • Core Specialty — Shepherd's original capacity provider (StarStone Specialty paper, excess casualty since Feb 2022) is itself a multi-line specialty insurer formed in 2020 with ~$900M of recapitalization. Frenemy risk in both directions: it sees Shepherd's book performance and could internalize the segment, or reprice/withdraw capacity in a severity cycle.
  • FM Global / Munich Re (data-center property specialists) — As Shepherd pushes from casualty into builder's risk and property for hyperscale data centers, it collides with the engineering-led property giants who have insured mission-critical facilities for decades and whose loss-prevention engineering is the incumbent version of telemetry underwriting.