Denied a claim? Insurf drafts the appeal, cites the policy, gets a physician to sign it, and tells you what the payer decided. Every case it resolves quietly builds something rarer than revenue: a map of how insurers actually behave.
A denied insurance claim is one of the most expensive pieces of paper in American medicine, and not because of what it costs to appeal. It is expensive because most of the time nobody appeals it at all. The letter arrives, a billing clerk sighs, the claim gets written off, and the revenue disappears into a number nobody reports. Insurf, a two-person company in Y Combinator's Summer 2026 batch, was built around a stubborn observation about that letter: it is usually wrong, and it is usually winnable.
The number the founders keep returning to comes from a KFF analysis of 2024 federal data. When Medicare Advantage prior-authorization denials get appealed, 80.7 percent of them are partially or fully overturned. Read that twice. The insurer says no, the practice pushes back, and four times out of five the no becomes a yes. The problem was never that the appeals fail. The problem is that they rarely get filed.
There is a reason for the silence, and it is not laziness. Appealing a denial is skilled, tedious work. Someone has to read the letter, find the right clinical documentation, match it against the payer's own coverage policy, write an argument, and route it to a physician for a signature - and then track whether the insurer ever responds. A busy practice generates more of these than any billing team can chase, so a triage happens by default: the biggest, easiest cases get fought, and the rest get written off. The write-off is invisible on any single claim and enormous in aggregate. Insurf's entire reason to exist is that the triage is a staffing problem, and staffing problems are exactly what software is good at absorbing.
Insurf's first product is called Inveto, and its job description is refreshingly narrow. It takes a denied claim and turns it into a source-cited, physician-attested appeal. Not a summary. Not a suggestion. A finished document that a doctor can sign and a practice can submit, with the relevant coverage policy cited inside it. The company's own tagline strips the pitch to four words: prior auth, denial appeals, end to end.
The "end to end" part matters more than it sounds. Insurf runs a six-step workflow that starts before the draft and ends after the decision. Intake pulls in the denial. The system drafts an appeal and cites its sources. A physician reviews it. Staff, not software, control the actual submission, so the practice keeps its hand on the button. Then Insurf follows up and reports the outcome. Crucially, it leaves the practice's existing systems intact - no rip-and-replace, no new EHR to learn.
There is a small design decision buried in that workflow that says a lot about the company. Insurf tracks the payer's decision separately from the posted payment. In revenue-cycle work those two things blur together, which is exactly how a "resolved" claim can quietly go unpaid. Splitting them is unglamorous. It is also the kind of thing you only build if you have watched the money leak.
The insistence on citations is the other tell. A generic language model can produce a fluent appeal letter in seconds, and it will confidently invent the policy language to support it. In healthcare, an appeal built on a hallucinated citation is worse than no appeal at all - it wastes a physician's signature and hands the payer an easy reason to uphold the denial. Insurf's requirement that every draft point back to a real, source document is partly a quality bar and partly a survival trait. It is the difference between a tool a compliance officer will approve and one they will quietly ban.
Insurf points at two kinds of customer, and it talks to each of them differently. For hospitals, the pitch is revenue protection on clinical denials - the high-dollar cases where a "not medically necessary" letter can erase a real procedure's reimbursement. For specialty practices, the pitch is prior-auth templates that speed the paperwork without taking the clinical decision away from the physician. That second promise is doing quiet work. Plenty of automation tools fail in medicine precisely because doctors will not cede judgment to a black box, and Insurf seems to know it.
The business model tracks the two products. Inveto is sold as software to providers, priced against the revenue it recovers, with a services layer of guarantees on top - the closer a vendor's pay is tied to claims actually overturned, the more its incentives line up with the practice. Surely, when it arrives, points at a different buyer entirely: employers and brokers who need to compare plans on true annual cost rather than sticker premium. One product earns its keep today; the other is a wager on owning data nobody else is collecting. Selling to both providers and payers-adjacent buyers is ambitious for two people, but it is also how the coverage-decision graph pays for itself twice.
The company also does something startups rarely do: it writes its own accountability into the contract. Miss a deadline, and Insurf credits the fee. Every eligible case is either worked or documented with a reason it wasn't. In an industry where vendors love to promise outcomes and disappear when they don't deliver, putting fee credits on paper is a differentiator you can actually enforce.
The bigger betHere is where Insurf gets more interesting than a better appeals tool. Every claim Inveto resolves teaches the company something specific: how a particular payer actually behaves when challenged on a particular kind of denial. Collected at scale, those observations become what Insurf calls the coverage-decision graph - a proprietary dataset of real payer behavior rather than published policy. Published policy tells you what an insurer says it will cover. The graph tells you what it actually does.
That dataset is the fuel for a second product, Surely. Where Inveto fights denials after the fact, Surely aims to price the true cost of a health plan before the fact, using cited source documents and version-pinned coverage policies to project a full year of out-of-pocket expense for employers and brokers comparing plans. It is a classic data flywheel: the first product is a business, and it also happens to generate the raw material for the second.
Turns denied claims into source-cited, physician-attested appeals. Six-step workflow, existing systems untouched, decisions tracked to outcome.
Prices the true annual cost of a health plan for employers and brokers, built on the coverage-decision data Inveto generates.
Whether the graph becomes a genuine moat is the open question of the whole company. Data flywheels look inevitable on a slide and turn out to be hard in practice - you need volume before the data is worth anything, and you need it before an incumbent with more claims notices the same opportunity. But the logic is sound, and it explains why a company that could have shipped a simple appeals bot chose to name its dataset instead.
Where it sitsDenial management and revenue-cycle automation are not empty fields. Large vendors have sold billing and claims software for decades, and a fresh wave of AI startups is now aiming at the same paperwork. Insurf's answer to "why you" is not scale - it has two employees - but specificity. It refuses to ship an appeal without citations. It keeps a physician in the loop. It writes accountability into the contract. And it is quietly hoarding a kind of data that generic drafting tools throw away.
Insurf's founders did not arrive from inside the insurance industry, which may be the point. Chang Lu, the CEO, came off a Brown University BS/MD track with a background in neuroscience and economics, and did zero-shot modeling research at Emory's Winship Cancer Institute before starting a company about claim appeals. Bryan Chung placed 18th internationally in a physics competition, holds a Platinum rating in USACO competitive programming, and has published twice in IEEE journals. Their YC partner for the batch is Gustaf Alstromer.
It is a technical pair pointed at a decidedly non-technical mess. That combination - people who could clearly build almost anything, choosing to build the appeals workflow instead - is its own kind of signal. The interesting startups are often the ones aimed at problems that look too boring to bother with, right up until someone does the arithmetic on how much money is quietly walking out the door.
What is worth watching from here is not whether Insurf can win an appeal - the 80.7 percent figure suggests the appeals are winnable when someone bothers - but whether it can do the boring part cheaply enough, at enough volume, to make the coverage-decision graph real before anyone else builds the same flywheel. The company is two people and pre-Demo Day, which means most of this is still promise rather than proof. But the shape of the bet is clear and unusually honest for a pitch: find the revenue practices are already leaving on the table, recover it with a document a doctor can stand behind, and keep the receipts. Everything else Insurf hopes to become is built on that first, unglamorous transaction repeated a few million times.