Insurance promises rescue at the precise moment ordinary life becomes paperwork. A collision, a damaged roof, a workplace injury - then come the calls, attachments, status requests, forms and repeated explanations. Cornelius Schramm has chosen this administrative afterlife as the place to build an AI company. It is an oddly practical destination for someone whose route passed through German-language machine-learning research, Division I tennis, fleet software and a Christmas coding sprint with graduate-school friends.
Schramm is the co-founder and CEO of Avallon AI, a New York company in Y Combinator's Spring 2025 batch. Its agents work across the unlovely trio of phone calls, emails and documents, translating them into structured claims workflows. The company calls its tools Claim Intaker, Case Status Receptionist, Third Party Reacher and Case Copilot. The names sound less like science fiction than an office directory. That is the point. Avallon is being built for work already happening, in systems insurers already use, with queues already waiting.
The wager is that an AI agent earns its place not by appearing clever but by carrying context from one awkward medium to the next. A claimant phones. A report arrives as a PDF. A vendor replies by email. Somewhere, a record must remain complete and an adjuster must know what to do. Schramm's career has repeatedly returned to this same transformation: messy input, specific action.
A thesis assembled from three clues
Avallon did not begin with a single cinematic revelation. Schramm has described three experiences that converged. His grandfather had been an insurance adjuster in Germany. At FINN, the car-subscription company, Schramm saw claims accumulate and watched the manual work around them. Then a roommate who worked as an adjuster supplied the view from inside: endless forms, phone tag and data entry, summed up as "endless drudgery." Family memory, operating experience and kitchen-table testimony all pointed toward the same bottleneck.
The founding team met while studying computer science and operations research at Cornell. During a fall-break hiking trip, they batted around startup ideas until Avallon began to take shape. They shipped a first version around Christmas, landed a customer weeks later and left graduate school to work on the company full-time. It is a brisk sequence, but the insight beneath it had been accumulating for years.
“We see our platform as a digital teammate, not a point solution.”Cornelius Schramm on Avallon's operating model
That distinction explains the breadth of the product. A document parser alone cannot complete a callback. A voice agent alone cannot reconcile a file. An email assistant alone cannot update the claims-management system. Avallon is attempting to join those actions into one operational thread. Schramm says the system also observes where processes stall, creating information that can improve the operation over time.
An anatomy of the queue
The operator before the founder
Schramm arrived at insurance automation by way of cars. In 2021, he joined FINN as the founding member of its U.S. automation team. The company needed applications that could coordinate a growing fleet, partner network, data model and changing set of business rules. Fifteen to eighteen fleet process managers were responsible for more than 20,000 cars. The job was less about polishing one permanent dashboard than about making software flexible enough to change every week.
Working with engineer Leander Peter, who would later become an Avallon co-founder, Schramm helped build FINN's internal operating hub. A central "Car Details" application consolidated subscriptions, vehicle transfers, partner communications and delivery timing. One tab that surfaced compound processes and estimated arrival times took Schramm two days to build. The wider setup grew to roughly 100 internal applications and saved hundreds of engineering hours.
This period supplies a useful explanation for Avallon's product taste. Schramm learned to begin with the people touching the process, follow their exceptions and then connect the scattered systems around them. At FINN he put it plainly: the ability to follow up with partners allowed the company to move a high volume of cars with very few employees. In claims, follow-up is again the hinge. A case can wait on one medical provider, one employer, one claimant or one attachment. Software becomes valuable when it remembers who owes what.
The operating playbook
- Start with a recurring queue, not a generic technology.
- Map the entire handoff across people, messages and systems.
- Build quickly enough to learn from the next exception.
- Keep humans on the decisions where judgment matters.
Specificity, before it became a company
In 2022, Schramm appeared among the authors of a paper on peer-to-peer feedback in higher education. The researchers asked whether software could identify how specific a sentence was, because specific feedback tends to be more helpful and instructors cannot manually check every comment in a large course. Using German-language student feedback, the team compared a conventional feature-based model with German BERT. The latter reached 81.1 percent classification accuracy; the simpler model reached 76.7 percent and retained lower computational costs and more interpretable coefficients.
A research paper about student feedback and a company processing insurance claims should not be forced into a tidy origin myth. Still, they share an intellectual habit. Both take language that arrives in human form and ask whether a machine can make it more legible, specific and useful. In the paper, the destination was helpful feedback. In Avallon, it is a claim record and an actionable next step.
Schramm's public record carries other hints of this builder-operator combination. He studied business administration at the University of St. Gallen, earned a master's degree in computer science at Cornell and conducted machine-learning research in Switzerland. At the St. Gallen Symposium, he helped manage more than 300 "Leaders of Tomorrow" participants and wrote Python scrapers to automate database maintenance. He also presented FINN's no-code story at a 2022 summit in Paris. The pattern is not devotion to one programming language or fashionable stack. It is a preference for removing coordination drag.
A company grows into the room
Avallon's early geography reads like a compact startup comedy: a borrowed room on Cornell's graduate campus, a cramped Airbnb during Y Combinator, then a coworking space in Williamsburg. In 2026 the team carried boxes into a loft in SoHo, its first dedicated office, after crossing seven figures in revenue. Schramm wrote that the new room had space for three times the team and immediately wondered how long that would last.
There was, briefly, a billboard phase. After YC, Schramm and co-founder Bryan Guin stood beneath an Avallon display in Times Square and took the obligatory photograph. The giant screen advertised "AI Agents for Claims Operations." It was a visible celebration of a business whose value is mostly invisible: a shorter queue, a completed callback, a document read correctly, a case that moves.
The $4.6 million seed round arrived in November 2025, led by Frontline Ventures with Y Combinator, 1984 Ventures, Liquid 2 Ventures and BOOOM participating. Schramm said Avallon's revenue had grown tenfold during YC and that its customers included administrators in the United States and Europe. The financing was earmarked for hiring and product development, with ambitions to expand from workers' compensation and automotive into broader property and casualty lines.
“Other companies automate tasks; we automate operations and learn from them.”Schramm on building across the full claims lifecycle
The culture between calls
The public version of Schramm is not all process maps. He played Division I college tennis at Nicholls State University and later served as president of the tennis team at St. Gallen. In company updates, sport repeatedly sneaks into the workday: team runs through the hills near Mount Diablo after customer workshops, a pause to watch Germany in the World Cup, an Avallon contingent at a charity 5K.
A 2026 offsite carried the team to Munich and then Erlangen, where Schramm grew up, for the hillside beer festival he had attended since childhood. Colleagues arrived in lederhosen and dirndls; his family joined; California customer calls continued between toasts. His own description was cheerfully economical: sometimes building company culture looks like "a bunch of Americans in a tiny German town. Because why not?"
That mixture of rigor and play may be the most human counterweight to Avallon's subject. Claims exist because ordinary days go sideways. The people processing them need reliable information, but they also need time and attention for judgment. Schramm's stated ambition is to let automation absorb repetitive work while adjusters focus on resolution. It is a modest-sounding proposition with a difficult standard: the software has to survive the disorder of real life.
For now, Avallon's story is still early and highly operational. The company is hiring engineers and salespeople, turning bespoke implementations into repeatable ones and trying to serve more customers per engineer. There are fewer philosophical manifestos than customer workshops. That suits Schramm's trajectory. He has spent years studying specificity, building internal tools and following the work until the actual bottleneck appears.
Insurance may never be delighted to answer the phone. But if Avallon works as intended, it will answer with the file open, the context intact and the next action already waiting. In claims, that may qualify as a small miracle of administration.