The AI Race Moved to the Construction Site. Shepherd Is Building the Insurance for It.

Justin Levine started his career as a civil engineer. He worked on large-scale commercial construction projects, fell into risk management by accident, ended up running the insurance P&L for a major New York City contractor through a captive, founded a construction risk management startup, sold it, and then spent 18 months at Autodesk running risk management strategy inside their construction technology ecosystem.

That last role is where the Shepherd thesis was born.

At Autodesk, Justin watched contractors tell him — anecdotally and empirically — that the technology they were using was making their job sites safer, reducing claims, improving quality. The data was there. The signal was clear. And the underwriting community was not using any of it.

There was a treasure trove of behavioral data about how builders operate in the field every single day. Data that did not exist 15 years ago because construction was a drawings-and-pen-and-paper industry. And nobody in insurance was touching it.

Shepherd was founded in 2021 to fix that. In Episode 162 of InsurTechTalk, Justin and I covered what that actually means in practice — how you underwrite a data center, why the Coalition-Allianz deal is a seminal moment for InsurTech, and why the construction site is now the most important insurance problem in the AI boom.

About Justin Levine

Justin Levine is the Co-Founder and CEO of Shepherd, an AI-native commercial insurance MGA focused on construction and renewable energy. He is a former civil engineer who ran risk management for a major New York City contractor, founded and sold a construction risk management startup, and led risk management strategy at Autodesk before founding Shepherd. Shepherd has raised $67 million in total funding, including a $42 million Series B led by Intact Private Capital, with participation from Spark Capital and Costanoa Ventures. The company has issued over 1,500 policies, covers over $400 billion in insured value, and serves more than 600 commercial contractors and asset owners.

The Problem: Speed, Not Capacity

The framing Justin uses for Shepherd's thesis is precise and worth understanding carefully.

Everyone assumes construction insurance for AI infrastructure has a capacity problem. Carriers want to write data centers, semiconductor fabs, and utility-scale solar. The demand for coverage is there. So what is the problem?

It Is a Process Gap, a Speed Gap, and a Data Gap

These projects are moving faster than any infrastructure build in recent memory — possibly faster than anything in several decades
A broker submitting a complex construction account to a traditional carrier can wait weeks for a response — weeks that a developer building a data center simply does not have
Submissions arrive as handwritten exposures, unformatted PDFs, messy spreadsheets — and get priced off data from three years ago that may have nothing to do with how that project is actually being run
The program does not have one carrier. It has dozens. Because each carrier is deploying less limit than they used to — $10 million today versus $25 million five years ago — building a $100–200 million casualty tower means syndicating across many markets simultaneously
In that environment, speed is everything. Brokers go to the market that answers first, prices fairly, and knows what it is doing. That is the competitive advantage Shepherd is building toward

Capacity Is Becoming a Problem Too

Justin was careful to note that while the primary gap today is speed and data, capacity is an emerging constraint — particularly on the property side, where insurable values on a single data center project can reach into the billions. The aggregation of all that value across the market is creating genuine limit pressure that will only intensify as the AI infrastructure build accelerates.

What Shepherd Actually Does

Shepherd is a commercial MGA writing primary casualty, excess casualty, and builder's risk for construction and renewable energy. It distributes exclusively through retail brokerage and operates largely in the admitted market — but trades like an E&S product because of the large commercial deregulation exemptions available at that scale, which give it freedom of rate and freedom of form.

How the Underwriting Actually Works

When a broker submits a construction account to Shepherd:

AI ingests the submission files — PDFs, emails, spreadsheets — and automatically structures the data before an underwriter has even opened the file
Shepherd can deliver an indication in minutes, not weeks
Beyond the submission, Shepherd integrates with the technology platforms contractors already use: Procore, Autodesk, DroneDeploy, OpenSpace
These integrations provide a real-time view of what is actually happening on the job site — incident management, inspection patterns, trade conflict tracking, quality documentation — not a static application from three weeks ago
The behavioral signals from these platforms feed Shepherd's predictive models, allowing it to price risk based on how a contractor actually operates, not just what class of business they are in

The $400 Billion in Insured Value

It is tempting to assume that $400 billion in insured value means Shepherd is primarily a data center insurer. Justin corrected that assumption directly: data centers are actually a minority of the portfolio. The majority is commercial building across sectors — healthcare, education, life sciences, heavy civil infrastructure (roads, bridges, tunnels). Construction is simply a high-spend industry, and insurable values across all of those sectors have been climbing steadily with material costs and project complexity.

The Behavior-Based Underwriting Revolution

This is the core of what makes Shepherd genuinely different — and Justin's analogy to personal auto telematics is the right frame.

What Happened to Personal Auto — and What Is Happening to Commercial Now

Twenty years ago, personal auto insurance was priced on simple demographic proxies. Then telematics happened. Thousands of data points replaced a handful. Algorithmic pricing replaced actuarial tables. Progressive, State Farm, and a few others figured out how to use behavioral data before anyone else did, and they built durable competitive advantages from it.

Commercial construction is going through the same shift — just 20 years later, and with a much larger and more complex data set.

The data did not exist before. Contractors used to run job sites on paper. There was no digital record of daily activity. The only way an insurance company could assess a contractor's behavior was to send a loss control engineer to do a single-point site visit and then extrapolate forward from one snapshot.

Today, platforms like Procore are used by nearly 20,000 contractors across the US. Every inspection, every incident report, every trade coordination log, every safety observation — all of it exists in a queryable digital record. Shepherd can plug into that record every day and see how a contractor actually operates, not how they described themselves on an application.

Shepherd Savings: The Most Differentiated Product in Construction Insurance

This program is worth understanding in detail because it solves a problem that has plagued prevention-based insurance products for years: the chicken-and-egg problem of technology adoption.

The Problem It Solves

Historically, contractors who invested in construction technology — drone reality capture, digital inspection platforms, quality management software — bore the full cost of that investment themselves. The benefit flowed largely to their insurer in the form of better loss performance, but the insurer never shared any of that value back with the contractor.

Shepherd Savings changes the equation:

Contractors who use qualifying technologies (DroneDeploy, Procore, Autodesk, and others) receive up to 25% lower premiums upfront
The savings frequently exceed the cost of the technology itself — creating a genuine win-win where the contractor's insurance cost funds their technology adoption
Shepherd monitors technology utilization through data integration throughout the policy term — not just at inception
Clients receive monthly reporting and feedback on technology usage and how it relates to their risk profile
Multi-year rate guarantees are available for contractors who maintain strong technology utilization

Why This Works When Wearables Did Not

Justin made an important distinction that connects directly to the conversation with Adam Price in Episode 155 about Kinetic Comp's pivot away from wearables. The core insight: Shepherd does not ask contractors to adopt new technology. It rewards them for technology they have already bought and are already using.

That eliminates the change management problem entirely. There is no political capital cost. There is no employee resistance. The contractor is already using Procore. Shepherd just asks for a data authorization agreement and starts underwriting the behavior the data reveals. The contractor's response, as Justin described it: "Why wouldn't we do this?"

Construction Meets Renewable Energy: A Natural Expansion

Shepherd's expansion into renewable energy was not a strategic pivot — it was a logical extension of what they were already seeing in the construction portfolio.

How It Happened

Almost every major data center being built right now includes its own independent power component — the grid cannot supply enough energy, so developers are building solar, battery storage, and other energy assets as part of the same development
Shepherd was already underwriting the construction phase of those energy assets as part of the broader data center project
The natural next step was to follow those assets into their operational phase — underwriting the renewable energy facility once construction was complete
Shepherd now has a dedicated energy practice with its own underwriting leadership, distinct from but continuous with the construction team

Justin also flagged what he sees as an emerging trend that will complicate and expand this further: developers building completely closed-loop systems — data center, power generation, battery storage, and water system all integrated into a single self-contained development. Effectively, building a small town. The complexity and insurable value of those projects is unlike anything the construction insurance market has historically had to handle.

Shepherd has identified seven distinct archetypes of data center construction, each with different risk profiles that require distinct underwriting approaches. The carriers that understand those differences will outperform the ones that treat it as a single class.

The Coalition-Allianz Deal: A Seminal Moment for InsurTech

This was the most forward-looking and arguably most important part of the conversation — and Justin brought it up unprompted, which tells you how much it matters to him.

What the Deal Means

Coalition is an MGU that built a proprietary cyber underwriting platform. Allianz — one of the largest insurance companies in the world — has effectively said: we cannot replicate what Coalition has built internally, so we are going to make them our exclusive underwriting partner for this line of business globally.

Justin's read on why this matters:

It is the first time a carrier of that scale has publicly acknowledged that a technology-native MGU can outperform their own underwriting capability
For years, carriers and reinsurers have been skeptical of the InsurTech thesis — "better data produces better underwriting results" — treating it as unproven marketing
The Coalition-Allianz deal is the first major proof point at institutional scale that the thesis is real
For a company like Shepherd that is making the same argument in construction and energy casualty, this is a north star moment — it validates the entire bet they are making

Justin was careful to note that cyber is different from construction casualty, and that Shepherd still has significant work to do to prove out its own results. But the direction of travel is now clear: the MGU model built on genuine data and technology advantage is not just a distribution play — it is an underwriting play. And carriers are starting to accept that.

"InsurTech Is Dead"

Justin pushed back on the idea that InsurTech as a category is finished. His view: the first wave of InsurTech was mostly about distribution efficiency and customer experience. Important, but table stakes. The second wave — the one Shepherd is part of, and that the Coalition-Allianz deal validates — is about underwriting performance. That is harder to build, takes longer to prove, but produces a fundamentally more durable competitive advantage.

The word InsurTech may be unfashionable. The underlying thesis — that technology produces better underwriting results — is just getting its first real proof points.

Key Takeaways

The AI infrastructure construction boom is creating one of the largest construction insurance opportunities in a generation — but the problem is process speed and data, not capacity
Behavior-based underwriting for commercial construction is where personal auto telematics was 20 years ago — the data now exists, and the carriers that use it first will build durable advantages
Shepherd Savings solves the prevention-adoption problem by rewarding contractors for technology they already use, rather than asking them to adopt something new
The Coalition-Allianz deal is a seminal moment for the MGU model — the first institutional validation that technology-native underwriters can outperform legacy carriers on loss ratio, not just distribution
Closed-loop data center developments — power, compute, water, storage in a single integrated system — are the next wave of complexity for construction insurers, and most are not ready for it
Admitted commercial markets at scale offer a structural advantage: freedom of rate and freedom of form, without the filing constraints that slow down smaller commercial lines