Picture a physician on a Tuesday afternoon, three patients behind schedule, staring at a fax machine that feels lifted from 1987. She's not treating anyone. She's arguing with an insurance company about whether it will pay for a drug she prescribed weeks ago.
Thirteen hours a week, on average. That's how much time U.S. physicians spend on prior authorization paperwork, per a 2024 American Medical Association survey. It amounts to more than a full workday each week spent convincing insurers to cover treatments already deemed medically necessary. For Chang Lu and Bryan Chung, the two founders behind Insurf—a pre-Demo Day startup in Y Combinator's most recent batch—those wasted hours look less like a problem and more like a market.
Their answer is Inveto, an AI-powered tool built to automate both the initial prior authorization submission and the appeals that often follow when insurers say no. The value proposition sounds almost too simple: feed the system a patient's electronic health record, and it spits out a complete, citation-backed submission in minutes. The company insists there are no hallucinations, no invented facts. The reason, they say, is that the system doesn't really "write" at all.
Facts, Not Fiction
Insurf's technical approach hinges on what Lu describes as "extraction, not generation." Rather than asking OpenAI or Anthropic's large language models to compose medical justifications from scratch, Inveto deploys them solely to identify and pull relevant facts from clinical charts. The system then assembles those extracted data points into prior authorization forms and appeal letters through what amounts to a deterministic process—every claim tethered to a specific line of text in the patient's record.
For prior authorizations, Inveto matches the correct insurance form template, auto-fills fields using data from the chart, and surfaces what the company describes as bound source quotes for a staff member to review. A physician signs off. The system exports a faxable document. (Yes, fax. This is healthcare.)
Appeals get more involved. The platform ingests the denial letter, the clinical chart, and the exact version of the payer's policy that was in effect at the time. It cross-references extracted facts against policy requirements and flags documentation gaps—places where the medical record doesn't quite line up with what the insurer demands.
Insurf claims a "0% hallucination rate, 100% source-cited accuracy," though these performance claims are presented as company marketing with no external validation available. One example on the Inveto product page shows an overturned denial for pembrolizumab, a cancer drug, that recovered $18,417 in 23 days by citing a specific UHC Medicare Advantage policy version. Whether that case is representative or cherry-picked is unclear.
Building a Map of Insurer Behavior
The immediate automation play is one thing. Insurf's longer-term bet is something else entirely: a proprietary dataset the company calls its "coverage-decision graph." Every prior authorization and appeal processed through Inveto logs a record—tracking real payer behavior across procedures, policies, and outcomes. The startup's pilot agreements, according to the company, include outcome-sharing provisions. That means it captures which denials get reversed, how long appeals drag on, and which insurers prove most stubborn.
This data strategy makes more sense when you consider the scope of the denial ecosystem. A January analysis from the Kaiser Family Foundation found that Medicare Advantage insurers issued nearly 53 million prior authorization determinations in 2024. When physicians appealed denials, roughly 80% were overturned. Yet only about 11.5% of denials actually get challenged—a staggering gap between what could theoretically be recovered and what doctors have the stamina to fight for.
Insurf's second product, called Surely, is a cost-prediction engine designed to simulate a member's true out-of-pocket healthcare expenses over 12 months. It accounts for premiums, deductibles, copays, coinsurance, drug tier structures, and out-of-pocket maximums. It also flags items likely to trigger prior authorization headaches. The company frames itself as "AI-native health insurance decision infrastructure," though both products remain in pilot phase. No public customer logos, no disclosed pricing.
The Competition Isn't Standing Still
Insurf is hardly alone in this space. The market for prior authorization automation has gotten crowded fast—perhaps more crowded than any two-person startup would prefer.
Innovaccer launched Flow Auth in September 2025, offering end-to-end AI-driven prior authorization that plugs directly into electronic health records and payer systems. Surescripts, an established health information network, piloted automation technology in 2024, with the latest updates in 2025. In May 2025, 1upHealth rolled out a FHIR-based electronic prior authorization solution aimed at health plans. Then there's Optum Rx's PreCheck tool, which reportedly cut prescription approval times from eight hours to 30 seconds, according to a report from earlier this year.
On the appeals side, the field gets even more crowded. Claimable announced an enterprise patient access platform in April. PrescriberPoint reported a 94.5% clinician acceptance rate for its AI-powered prior authorization agent around the same time. Counterforce Health, a consumer and provider appeals assistant, drew coverage from Axios last August. Smaller players like Revguard, InvisaClaim, and MedAppeals are also in the mix. And on the payer side, Cohere Health raised $90 million in May 2025 to expand its clinical intelligence and automation platform.
Regulatory shifts may provide some lift. CMS's Interoperability & Prior Authorization Final Rule requires payers to start exposing prior authorization metrics—denial rates, turnaround times, volume—beginning in 2026, with API requirements for electronic prior authorization rolling out in January 2027. That transparency could, in theory, make Insurf's coverage-decision graph more valuable. But only if the startup can capture enough volume to identify meaningful patterns before competitors do the same.
Two Founders, One Big Bet
Lu, the CEO, studied neuroscience and economics at Brown and conducted zero-shot modeling research at Emory's Winship Cancer Institute. Co-founder Chung has a background in competitive physics and computer science. The Y Combinator directory lists the team size as two, though LinkedIn shows a broader company size band of 2-10 employees—a range captured in late June, which may or may not still hold.
Insurf is currently offering a 90-day Access Recovery pilot aimed at specialty practices. The terms include outcome-sharing rights and no upfront IT approval requirements, a pitch designed to lower friction for potential partners. The company's homepage notes its pre-Demo Day status, signaling that a more formal public launch is close but hasn't quite arrived yet.
The Weight Physicians Carry
The problem Insurf is addressing is real, even if the solution remains unproven at scale. The AMA's 2024 survey found that 94% of physicians reported prior authorization delays in necessary care. Roughly one in four said those delays had led to a serious adverse event. Nearly 80% reported patients abandoning treatment entirely due to prior authorization struggles. Physicians complete an average of about 40 prior authorizations per week—on top of those 13 hours of administrative time.
A research paper published in March 2026 on AI-generated prior authorization letters noted that while large language models produce "strong clinical content," they often deliver "weak administrative scaffolding." Meaning the medical reasoning is sound, but the structure and completeness fall short of what payers actually require. Insurf's extraction-first approach and deterministic assembly seem designed to sidestep exactly that failure mode. Whether it works in practice, across edge cases and constantly evolving payer policies, is a different question.
The company's challenge isn't proving the problem exists—physicians and health systems feel it acutely, every single day. The challenge is proving differentiation in a field where multiple well-funded competitors are shipping similar promises, often with larger teams and deeper pockets. Whether Insurf's coverage-decision graph becomes a genuine competitive moat or just another dataset in an increasingly crowded market remains an open question.
For now, it's two people with a clear thesis, a working product, and a 90-day window to prove the model scales. That's the bet. Whether insurers—and the fax machines that still somehow define their workflows—are ready for it is another matter entirely.
