I've spent most of my career building insurance companies and working with the people who build them. One thing has stayed true that entire time: almost everyone has significant ambition for what technology could do for their business, and very few have the time, expertise, and capital it takes to realize it.
Right now, the clearest example is the MGA channel. The segment has grown from $47 billion to $114 billion in written premium since 2020, yet fewer than 7% of MGAs have AI agents working in production. [Source: The Specialty MGA Operating Model Inflection Point.] It would be easy to read the lack of production AI as caution, or a lack of appetite. I read it differently.
Every MGA operator already knows submission triage is slow, that bordereaux reconciliation eats a disproportionate share of ops headcount, that claims intake could move faster than it does. The people best at underwriting, distribution, and servicing are spending their days underwriting, distributing, and servicing, which is exactly what they should be doing. Standing up a technology function from scratch takes a year or more and a seven-figure hiring spree, and most MGAs would rather spend that time and money growing the book. Correctly, in my view.
I've seen what that tradeoff costs firsthand. We spent a decade at Clearcover investing in proprietary technology to run our business. We built it ourselves, and it worked: by early this year, more than 90% of our claims intake ran through AI agents, and 93% of our policies were bound digitally. The work was also nothing like a straight line. We built things that didn't pay off, not because they failed, but because the ROI wasn't there. Those lessons about where to invest shaped how we operate just as much as our wins did.
One of the main lessons is that the hard problems are workflow problems, not model problems. A submission arrives missing three data points: what happens next? A servicing request touches four systems that don't talk to each other: who reconciles it? A claim needs a coverage decision at 11 p.m. on a Saturday: does anything move before Monday morning? Can we accelerate the workflow using AI?
Answers to those questions come from people who have sat inside an underwriting or claims operation, redesigned the work, and then built the system that runs it. They do not come from a better model, and they definitely don't come from a slide deck.
That points to a different shape of help than this industry usually gets offered. What closes the gap is a build partner: people who bring the insurance operating experience and the pre-built technical foundations with them, stand the system up inside the MGA's actual workflow, and leave the MGA owning it when they're done.
That is a different deliverable than a platform license the MGA has to configure and hope fits, and a very different one than a strategy roadmap. The MGA channel has spent decades outsourcing specialized functions to people who show up with the expertise already built, from actuarial to claims administration. Technology should work the same way.
The economics have to change shape, too. If adoption requires an MGA to write a large check before seeing results, most of the market will rationally pass, including plenty of the best-run shops. Adoption moves when the risk sits with whoever is doing the building and payment is tied to outcomes the business already counts: submissions triaged, servicing events completed, claims closed. Insurance has priced plenty of vendor relationships this way for decades. The approach just hasn't been applied to AI infrastructure yet.
We believe that enough to test it with our own capital. This month, we opened applications for Launchpad, a program where we fund and build the AI infrastructure for a small cohort of MGAs, the MGA owns what we build, and we earn as it produces results. I don't know yet whether this exact model is the one that closes the gap. I do know that the ambition in this channel was never the problem and that the operators who are best at this business shouldn't have to become something else to get the technology their book deserves.
