The New York startup building AI agents for the least glamorous corner of insurance - claims - just raised $4.6M to run the whole back office, not fix one step of it.
Ask anyone who has filed an insurance claim, and the story rhymes. A phone tree. A voicemail. A form re-typed by hand. A promised callback that arrives, if it arrives, a week later. Cornelius Schramm has a blunter way of describing it: the claims process, he says, was not just slow. It was systematically broken.
Schramm is the co-founder and CEO of Avallon AI, a New York company that came out of Y Combinator's Spring 2025 batch with an unfashionable idea. While much of the AI world chased chatbots and image generators, Avallon pointed its agents at the paperwork nobody wants to do - the intake calls, the medical reports, the endless back-and-forth of an insurance claim. The pitch is simple to say and hard to build: let software run the back office.
A single claim can involve a dozen phone calls and a stack of documents that never quite match. An adjuster spends the afternoon dialing an employer to confirm a detail, then a body shop to check a repair estimate, then re-keys the answers into a claims management system built sometime around the fax era. Multiply that by a caseload, and the math gets grim.
The timing, Avallon argues, is not an accident. The industry is staring at a staffing cliff - by some estimates, roughly 400,000 insurance workers are projected to leave the field by 2026 - just as large language models became good enough to hold a real conversation on a phone line. Schramm frames the moment plainly.
"The convergence of staffing challenges and advanced LLMs creates a unique opportunity for AI-native automation in insurance claims management."
Cornelius Schramm, Co-Founder & CEOMost "AI for insurance" tools do one thing - summarize a document, or answer a FAQ. Avallon's bet is that a claim is a workflow, not a task, so it ships a whole roster of specialized agents that hand work to one another. They have job titles, not model names.
Captures new losses by phone, email, or file upload - with zero manual data entry.
Answers inbound status calls with real-time updates and full claim context.
Auto-dials employers, providers, and injured workers, then logs every response.
Retrieves data, cites its sources, analyzes exposure, and suggests next actions.
Underneath sits the plumbing: a Dialer to scale inbound and outbound calls, an Orchestrator for email, and a Parser that turns messy PDFs, photos, and reports into structured fields. The agents plug into the systems insurers already run - claims management platforms, IVR phone systems, and data warehouses - rather than asking anyone to rip and replace.
Avallon sells to the companies that live and die by claims throughput: insurance carriers, managing general agents, and third-party administrators - the TPAs that process claims on behalf of everyone else. Its named customer, Athens Administrators, is a California-based multiline TPA. One partner runs a claims floor of more than 400 adjusters.
"Avallon's platform has made it easy to integrate AI into our operations quickly and effectively."
Danny Smith, VP IT, Athens AdministratorsThe traction shows up in the numbers. Avallon reported roughly tenfold revenue growth during its YC run and has reached six-figure ARR - modest in absolute terms, but a steep curve for a company founded the same year.
Avallon's pricing is built to survive a skeptical buyer. It starts with an upfront implementation fee and forward-deployed engineers who sit close to the customer's operations. After that, the meter runs on variable terms - fixed volume pricing, flat commitments, or, notably, pay-per-successful-outcome. You can pay Avallon when a claim is actually resolved.
That structure does something subtle in an industry that has been burned by software promises: it puts Avallon's revenue on the same side of the table as the customer's results. If the agents do not close the loop, the invoice reflects it.
Schramm was the founding US engineer at FINN, a car subscription startup where he built the fleet operations platform and scaled it into the millions in revenue; the company was later valued north of $650 million. He studied computer science at Cornell and did machine learning research in Switzerland. His co-founders - Bryan Guin, Moritz Bartusch, and Leander Peter - bring their own mix of Cornell, HSG, and MIT, with prior stops at FINN, EY, Taktile, and, tellingly, the insurer Allianz.
It is a team that seems to genuinely like unglamorous industries - the kind where the software is old, the workflows are ugly, and the leverage is enormous precisely because so few founders want to look.
"By using voice AI to understand claim-specific context and embedding into workflows, Avallon is positioned to deliver profitability gains for forward-thinking TPAs."
Will Prendergast, Frontline VenturesThe competition falls into a few buckets: point-solution insurtech tools that automate a single step, the big claims-outsourcing shops that throw people at the problem, and horizontal voice-AI and automation vendors that know nothing about insurance. Avallon's wager is that owning the entire claims operation - and understanding it deeply enough to price by outcome - is harder to copy than any one feature.
With $4.6 million from Frontline Ventures, Y Combinator, 1984, Liquid2, and Booom now in the bank, the company has room to test that thesis against real caseloads. The insurance back office is not a place that rewards hype. It rewards claims that close. That, for now, is the whole game.