A $500B industry still moves at the speed of retyping
Picture a commercial insurance broker on a Tuesday morning. A trucking company wants coverage for its fleet, and the broker has everything: the client's operations, the drivers, the vehicles, the loss history. Now that information has to reach the carriers who might write the risk, and every carrier wants it in a different shape. One takes an ACORD form. Another has a portal. A third wants its own supplemental application plus five years of loss runs, and a fourth will only look at the account if the driver schedule arrives in its preferred format.
So the broker starts typing. The same business name, the same addresses, the same vehicle identification numbers, entered again and again into PDFs and web forms that will never talk to each other. Brokers who handle complex risks spend roughly half their working day on this kind of re-entry. Not advising clients or winning new accounts. Retyping.
Commercial insurance is a $500 billion industry, and this is how most of it still runs: email threads and PDF attachments, keyed in by hand at every step. The submission workflow, the process of getting a client's risk in front of carriers, is the single largest bottleneck in how insurance gets distributed.
That bottleneck is what Casey exists to remove.
Casey builds AI agents that automate the submission workflow for commercial insurance brokers. The platform ingests client and risk data from the systems brokers already use, fills out carrier applications and supplemental forms automatically, assembles complete submission packages, and then handles the follow-up questions that come back from underwriters. Upload the client's information once, and Casey produces what every market in the deal needs. The company went through Y Combinator's Fall 2025 batch and is based in San Francisco.
What a submission actually takes
From outside the industry, "submitting a risk to a carrier" sounds like sending an email. Inside it, a submission is a document package with real assembly costs.
A complete submission for a mid-sized commercial account typically includes ACORD forms, the standardized applications the industry runs on, plus whatever supplemental applications each carrier layers on top. For anything involving vehicles, there are driver lists and vehicle schedules. For any account with history, there are loss runs, the claims records that tell an underwriter how the risk has actually performed. Each document draws on the same underlying facts about the client, and each has to be produced separately, formatted to the standards of whoever is receiving it.
The data to fill these documents already exists. It sits in the broker's agency management system, in the client's records, in prior policies, in the loss runs themselves. The work is not discovery. It is transcription, done by licensed professionals whose time is worth far more than the task, with every keystroke a chance to introduce the kind of discrepancy that stalls a deal weeks later.
And speed matters more than it might seem. When a business goes to market for coverage, several brokers may be competing for the same account, and the one who gets clean submissions in front of the right carriers first often controls the deal. When assembly takes days, the workflow itself decides who wins.
One upload, every form
Casey's product attacks the transcription layer directly.
A broker brings in raw client and risk data, from their existing systems or from the documents the client provides, and Casey extracts and normalizes it into a structured risk profile. From that single profile, the platform generates the paperwork: ACORD forms filled out, supplemental applications completed, driver lists and vehicle schedules formatted, loss runs organized into a package an underwriter can actually evaluate. Where outside data helps build the picture of the risk, Casey enriches the profile with sources like telematics. The output is a carrier-ready submission for any market the broker wants to approach.
The platform also reads the documents it touches. Because it holds the full risk profile, it can compare what a policy or an application says against what the underlying data shows, flagging coverage gaps and discrepancies that would otherwise surface as problems after binding. That is the kind of checking a careful broker does when there is time, and there is rarely time.
None of this changes what a broker sells. It changes how much of the day goes to producing documents versus advising the client. Casey's founders describe the end state as a brokerage where the humans focus on client relationships and closing business while AI handles the operational work behind the scenes. The submission workflow is the first and heaviest piece of that operational load, which is why they started there.
The follow-up loop
A submission does not end when the package goes out. Sending it opens a correspondence.
Underwriters come back with questions. Some are requests for information that was in the package all along, just not where the underwriter looked. Some genuinely require going back to the client. In a manual workflow, every one of these lands in the broker's inbox and joins the queue behind every other account, and each round trip can add days to a deal that was urgent when it started.
Casey runs this loop with a multi-agent system. When a carrier inquiry arrives, the agents compare what the underwriter is asking for against the data Casey already holds. Questions the existing profile can answer get answered from it. Only the genuinely missing pieces turn into a request to the client, and Casey drafts that request. The broker stays in the loop for judgment calls, but the mechanical work of triaging inquiries and chasing details stops consuming their day.
This part matters because follow-ups are where submissions go to die. A package that took days to assemble can sit for another week because a two-line question landed while the broker was underwater. Compressing that loop compresses the whole distribution cycle, and that is what brokers are actually buying: deals that keep moving.
Founders who watched the bottleneck up close
Casey's three founders earned this problem the direct way, at grape insurance, Switzerland's leading insurtech.
Pascal Küng was grape's first employee and led its engineering. Nico Hänggi was employee number two and a tech lead. Maximilian Thoelen joined as the tenth employee and served as chief of staff. Küng and Hänggi both studied computer science at ETH Zurich. At Casey, Thoelen is CEO, Küng is CPO, and Hänggi is CTO.
Building and scaling an insurtech from the earliest days means living inside insurance operations rather than reading about them. The trio watched how information actually moves between clients, brokers, and carriers, and where it grinds to a halt. Their conclusion was that the industry's distribution machinery, the whole apparatus by which risks find capacity, was the layer worth rebuilding, and they started Casey in San Francisco to rebuild it.
The founder-market fit here is specific. Insurance punishes outsiders who assume the paperwork is dumb. ACORD forms, loss runs, and supplemental applications encode decades of underwriting practice, and a tool that gets the details wrong creates more work than it saves. A team that has operated inside the industry knows which conventions are load-bearing, and that knowledge is hard to shortcut.
The infrastructure layer thesis
Casey describes itself as building the infrastructure layer for insurance distribution, and the framing is worth taking seriously.
Distribution in commercial insurance is the connective tissue between a business that needs coverage and the capital willing to provide it. Today that tissue is made of people re-entering data. Every broker-to-carrier handoff is a format conversion done by hand, which is why the industry's information moves at typing speed even though every party involved runs modern software internally.
An automated submission layer changes the economics of the handoff. When turning a risk profile into any carrier's required format costs nearly nothing, brokers can approach more markets per account, carriers see cleaner and more complete submissions, and the match between a risk and the right capacity gets better on both sides. The submission product is the wedge, but the position it establishes, sitting on structured risk data as it flows from client to market, is the platform.
The pattern rhymes with what happened elsewhere. Other industries had their manual handoff layers absorbed into software, and the companies that did the absorbing became infrastructure rather than tools. Insurance distribution has resisted this longer than most, partly because the workflows are genuinely complicated and partly because the incumbents' revenue is not hurt by the friction. A startup carrying no legacy book has no such attachment.
The quiet half of the day
There is a version of AI-in-insurance that promises to replace underwriting judgment or price risk with a model. Casey's bet is more grounded and, for brokers, more immediately useful: give back the half of the day that goes to retyping.
It is automation the people being automated can root for. Casey does not compete with the broker; it removes the part of the job no broker wanted. The relationships, the market knowledge, the negotiation, the judgment about which carrier fits which risk all stay human. The transcription goes to software, where it should have gone years ago.
Commercial insurance will still run on trust and expertise. Casey is betting it no longer has to run on data entry, and that the company which automates the industry's paperwork ends up holding the rails the whole market moves on.
