Why Launchpad

We spent a decade building AI and its precursors inside a live carrier at Clearcover, across claims, underwriting, rating, distribution, and service. That work taught us where AI actually pays off, and it also showed us how much time and capital a great AI build takes.

When we launched Dearborn Labs to bring best-in-class AI technology to the insurance industry, we looked hard at the MGA market. Great MGAs are exceptional at running a lot of business with a lean team, but running the shop is all-consuming, and there is rarely time or money left to build great AI. The build is expensive and slow, so the work that would compound an operator's advantage keeps getting deferred.

We think that is a bad trade for operators this good.

So we built Launchpad, an AI Insurance Accelerator by Dearborn Labs, to put the infrastructure we already built behind their businesses. We fund and build the AI infrastructure and workflows for a small cohort of MGAs, the partner owns what we build, and the partner pays only when it produces results. We designed it so we win only when you win.

That's the whole program in a paragraph. The rest is why we set it up this way, and why the structure matters more than it looks.

Why it matters to MGAs

Start with what everyone in this arena is chasing. Submissions, quoting, endorsements, claims handling: a growing set of well-funded AI platforms and dev shops are going after the same operating pain, and much of what they build is genuinely good. We compete for the same problems. Where we part ways is the business model underneath the demo.

Most of them sell a workflow product you subscribe to. You rent a shared workflow and inference layer. But so does your competitor down the street. Whatever edge that platform creates, everyone on it gets the same edge. Over time it stops being an advantage and settles into a cost of doing business. You pay for it forever, you never own the asset, and the vendor's incentive is to keep you renting.

The customization answer doesn't change this. Most platforms tailor themselves to each customer now. But customizing their workflow on their platform isn't the same as owning yours. The core still belongs to them. You're still renting, and the more they tune the system to exactly how you win, the more of your edge ends up encoded in a product that serves your competitors next quarter.

Owning the build means owning the workflow, the data, and control over the inference. That conversation is starting to surface across the industry, and we don't think it's going away.

What makes Launchpad different

We are a services provider, not a product to buy.

We build software the MGA owns, designed around how that specific shop runs, rather than a shared layer rented by everyone.

Of course, owning it doesn't mean we hand you code and disappear. Maintenance is planned and staffed as part of the engagement, so what you own stays supported. Remember, we succeed when you succeed.

Launchpad is built on the belief that MGAs should seriously weigh the build versus buy equation since the formula is changing rapidly.

When an MGA chooses to build today, the default answer is hiring a big-budget consulting firm, the kind of build that takes years, literally. We were talking to a client that was quoted a four year build, compared to our conservative eight weeks.

That's the slot we step into, and we're built for it in a way neither those firms nor the platform vendors are, on three counts:

• •We ran that carrier in production for a decade, with real claims still running through it today. Those consulting firms bill hours to learn insurance, and most of the AI firms here learned it from the outside in. We're insurance operators who build AI, not the reverse.
• •We start every build from a library built to run that carrier, not built to sell. Much of what a workflow vendor packages as its product already sits in that library as a starting point: submission intake and triage, document and bordereau ingestion, quote and rating flows, endorsement processing, a claims co-pilot, policy admin, compliance and QC, and the observability layer that tells you when a model is actually right. We don't start at zero. The library is the scaffolding; your workflow, your data, and your edge live in the build you own and never feed back into it.
• •We build AI-native. We haven't hand-written production code since early this year, so a build that used to take years and a fortune now ships in weeks. That speed is the whole reason build-versus-buy is a real choice now instead of a slogan. You end up with a system you own instead of a subscription you never stop paying.

The proof is in what that library already does. Our claims co-pilot earned 92% voluntary adoption across a 25-person front-line claims team and returned 617 hours a month, a productivity lift equal to about 4.7 full-time roles of research and documentation work. On the servicing side, expense as a share of earned premium fell from 4.2% to 2.3%, roughly a 45% reduction, while customer satisfaction rose and about 47% of more than 17,000 monthly contacts now resolve without reaching a human. That gives a team back the hours where judgment actually matters, and it grows what a lean shop can take on without adding headcount.

One clarification, because you see the cheap-proof-of-concept and pay-us-after-we-build models constantly: Launchpad isn't that. These are full production builds we'd normally charge into the hundreds of thousands for. That's our own capital at risk, which is why we vet and select a small cohort rather than opening the doors. We pick builds we believe will ship and perform.

We build it. You own it. That's the contract term, not a slide, and it is the same answer we would want a partner to give us.

If you run an MGA and you're weighing one of these decisions, applications for the inaugural cohort are open through September 8, 2026. The next step is a conversation, not a commitment.

Kyle Nakatsuji is the founder of Dearborn Labs and CEO of Clearcover, where the team has built and run production AI for nearly a decade.