ISSUE 06 · 8 MIN

Health insurers selling on the federal marketplace denied about 85 million in-network claims in 2024, and patients appealed 262,982 of them, which is fewer than one in a hundred. The business runs on that gap. Holden Karau, Neal K. Shah, and Warris Bokhari built agents that write the appeal in minutes for nothing, or for $39.95, on terrain where the old price for help was 15 to 35% of whatever the insurer finally paid.

A claim denial is a letter that says the insurer will not pay for something a doctor already decided you needed, and it is written so that the cheapest response is to give up. To fight it you have to find the plan document, work out which clause the denial leans on, find the clinical guideline or the published study that says the treatment is standard care, write all of that into a letter in the insurer's own vocabulary, and send it to the right address inside a deadline that is usually 180 days and sometimes far less. Most people never start, and the ones who do usually stop at the first page.

The builds in this issue give that job to an agent. You photograph the denial letter, the software reads it, asks a handful of questions about the treatment and the plan, pulls the medical evidence, and returns an appeal letter that cites the plan language, the state or federal rule that applies, and the literature, in the form a claims reviewer expects. Fight Health Insurance does that for whatever you choose to pay, including nothing, and offers a fax service so you never touch a fax machine. Counterforce Health does it free and says it always will. Claimable does it for $39.95, mails or faxes the letter for you, and sends reminders while the clock runs. Between the first two alone, more than 20,000 appeals have gone out that would otherwise have been a shrug and a credit card bill.

The problem these builds attack is not that appeals fail. It is that the system is priced on the assumption that nobody files one. KFF's analysis of the 2024 federal marketplace data found that insurers denied about 19% of in-network claims, with individual insurers ranging from 3% to 36%, and that of the roughly 85 million denials, consumers appealed 262,982, which rounds to zero. When a patient did appeal, the insurer reversed itself about a third of the time. The number is higher where the data is better: in Medicare Advantage, insurers made 52.8 million prior authorization decisions in 2024 and denied 4.1 million, only 11.5% of those denials were appealed, and 80.7% of the appeals that were filed won. The federal inspector general looked at a sample of Medicare Advantage denials in 2022 and found that 13% of the denied requests met Medicare's own coverage rules. A denial, in other words, is a bet that you will not come back, and the insurer wins that bet about ninety nine times in a hundred.

The bet is engineered. ProPublica reported in 2023 that Cigna's PXDX system let a single medical director reject claims in batches, about 300,000 in two months, at an average of 1.2 seconds per claim, without opening the file. The Senate Permanent Subcommittee on Investigations reported in October 2024 that UnitedHealthcare's prior authorization denial rate for post-acute care in Medicare Advantage went from 10.9% in 2020 to 22.7% in 2022, that Humana's denials for long-term acute care rose 54% in the same period, and that UnitedHealthcare had explored machine learning to predict which denials would be appealed. The estate of a Minnesota man is still in federal court against UnitedHealth over the nH Predict algorithm the company used to cut off rehabilitation coverage, with the case scheduled into this autumn.

The people who would do it for you charge like lawyers. A medical billing advocate typically takes 15 to 35% of the money recovered, and patient advocates bill by the hour. Hospitals, which cannot walk away from a denial the way a patient can, spent $25.7 billion in 2023 arguing with insurers over claims, up 23% in one year, at an average of $57.23 per claim, and Premier estimates that $18 billion of that was spent overturning denials that should have been paid the first time, because about 70% of them eventually were. Experian's 2025 survey found that 90% of denied claims need a person to rework them before they can be resubmitted, and that rework is the toll, paid in salaried hours on the provider side and in surrendered claims on the patient side.

The incumbents know which way this is going. UnitedHealthcare announced on 5 May 2026 that it will drop prior authorization for a further 30% of the services that still require it by the end of this year, said that roughly 92% of the authorizations it receives are approved within a day, and is expanding a gold card program that exempts high-performing practices from review entirely. The AHIP pledge signed by the largest insurers in June 2025 had cut prior authorization requests by 11%, about 6.5 million fewer requests, by April 2026, with a commitment to real-time electronic decisions by 1 January 2027. California's SB 1120 has required since January 2025 that a licensed clinician, not an algorithm, make any denial based on medical necessity. UnitedHealth Group booked $447.6 billion of revenue in 2025 and guided to more than $439 billion for 2026 with a medical care ratio near 88%, which means the margin it defends is roughly twelve cents of every premium dollar, and the cost of an appeal was one of the things holding that margin in place. The guerrilla's price advantage sits at the point where the marginal cost of a well-written appeal falls to the price of a model call, because a bet that nobody will file only pays while filing is expensive.

A mention here is listed, never sold.

LISTED · OPEN SOURCE, PAY WHAT YOU WANT

Fight Health Insurance

Upload the denial, the software reads it (with an on-device option so the medical details never leave your machine), and it drafts several appeal letters that cite the plan and the evidence, which you send yourself or through the platform's fax service. The site reports more than 10,000 appeals generated across all 50 states, and the whole thing is open source under the totallylegitco GitHub organization, built on Django and Ray with several model backends. Pricing is pay what you want, with $0 accepted. A provider-facing version, Fight Paperwork, launched in June 2025. Built by Holden Karau, the Apache Spark committer, with co-founder Melanie Warrick.

LISTED · FREE FOR PATIENTS

Counterforce Health

Upload the denial letter, the plan details, and the medical records, and the agent builds the case from peer-reviewed literature, clinical guidelines, and the regulations that apply, then returns the appeal in about two minutes on average. The site reports more than 10,000 appeals generated and claims a 75% approval rate against a national baseline it puts at 37.5%; treat both as the builder's numbers. No venture money: grants from the NIH and the University of Pennsylvania's innovation programs and an impact fund, and a stated promise that individuals will never pay. Founded in Durham by Neal K. Shah with Gavry Eshet, Riyaa Jadhav, and Maggie Xu.

LISTED · $39.95 PER APPEAL, FILED FOR YOU

Claimable

The fully packaged version. The patient uploads the denial and the insurance details, answers questions about the condition, and Claimable writes an evidence-based appeal, mails or faxes it to the right recipient with an expedited option, and sends reminders through the process; it does not handle prior authorization or external review. Flat $39.95 plus shipping, no success fee, and free appeals for many medications through agreements with four drugmakers, which is how the company acquires patients without paying for ads. Claimable reports that about three appeals in four succeed and that most are resolved within ten days. About $10 million raised, Mark Cuban among the investors. Founded by Warris Bokhari, a physician, with Zach Veigulis, formerly chief data scientist at the Department of Veterans Affairs, and Alicia Graham.

Take one appeal at Claimable's price and run it the way a CFO would. The patient pays $39.95. Card processing at Stripe's published rate of 2.9% plus 30 cents takes $1.46, so the builder nets $38.49 before doing anything. None of the three builders publishes what an appeal costs them to produce, so the figures that follow are assumptions, and deliberately generous ones: say the model calls to read the denial, search the evidence, and draft the letter cost $2, the document handling and fax or postage cost $3, and the support time averaged across every appeal costs another $5. That is a cost to serve of about $10, which is the term for everything you spend delivering one unit after the sale is made, and it leaves a contribution margin of roughly $28, or about 70% of the price. Contribution margin is what each sale contributes toward the fixed costs of the business, the salaries and the compliance work and the rent, before those costs are counted. A business with a 70% contribution margin and a $40 price needs volume, not price increases, and the market makes 85 million denials a year.

Now look at the same transaction from the customer's side, because that is what decides whether the volume shows up. Suppose the denied bill is $1,000. In the marketplace data about one appeal in three succeeds, so the expected value of filing, the probability of winning multiplied by what you win, is about $330 for a $40 outlay, and Claimable and Counterforce both claim rates far above one in three for the conditions they cover. A billing advocate working the same $1,000 denial on contingency would keep $150 to $350 of the recovery, and an hourly advocate could bill past the size of the claim before the letter was sent. The old market failed on both ends: for anything under a few thousand dollars the advocate's fee ate the recovery, and for the patient the time cost exceeded the expected value, so the rational move was to pay the bill. A flat $40 with a one-in-three floor and a claimed three-in-four ceiling makes the appeal rational for almost every denial larger than a copay.

The part worth copying is how Claimable pays for customers. Customer acquisition cost, the money spent to win one paying customer, is the number that kills most consumer businesses at this price point, because you cannot spend $60 on ads to sell a $40 product. Claimable's four agreements with drug manufacturers, under which the manufacturer covers the appeal for patients denied its medication, move the acquisition cost onto a party that earns thousands of dollars per patient who stays on therapy. The manufacturer's incentive and the patient's incentive line up, and the builder sits between them collecting a fee from whichever side values the appeal more. For the business to work at scale, three things have to hold: the win rate has to stay high enough that word of mouth replaces advertising, the cost to serve has to keep falling as model prices fall, and the insurers have to keep denying, which every number above says they will.

The version you can build by the weekend is narrower than any of the three above. Pick one condition, one drug, and one insurer, ideally a combination where a patient community already exists online and complains about the same denial letter in the same words. Clone the Fight Health Insurance repository, keep its structure, swap in the model provider you have a business associate agreement with (that agreement is the HIPAA contract that lets a vendor handle protected health information, and you cannot send medical records through a consumer API without one), and hard code the plan language, the guideline citations, and the appeal address for that single combination, so the agent has almost nothing to get wrong. Charge a flat $29 per appeal, or offer it free and charge the practice: the physician's office that treats that condition is drowning in the same denials, spends 13 hours a week on prior authorization according to the AMA's survey, and will pay a flat monthly fee for a tool that produces the letter from the denial and the chart note. The first customer is the billing person at a two-physician specialty practice, and the sale is a fifteen-minute demo on one of their own denials. Add fax delivery through any API that offers it, log every letter you send with the date so the deadline math is provable, and keep the patient's records only as long as the appeal is open.

totallylegitco/fighthealthinsurance is the repository behind Fight Health Insurance: a Django application with Ray workers, PostgreSQL, Docker and Kubernetes manifests, a React front end, and adapters for several model providers, with 1,409 commits, 154 stars, and 44 forks at the time of writing. Read the license file before you ship anything on top of it. It is worth an hour because it is the only place you can read, end to end, how a working appeal agent handles the ugly parts: reading a scanned denial, keeping medical details on the client where possible, choosing among several drafted letters, and sending a fax, none of which appears in a tutorial.

Here is how this one gets gamed, from the side of the desk that used to investigate it. An appeal service asks for exactly the bundle an identity thief wants: the member ID, the date of birth, the diagnosis, the treating physician, and the medical records, all volunteered by a frightened person who has just been told no. A copycat site with a plausible name and a $30 price collects that bundle and never sends a letter, and medical identity theft is slower to detect and harder to unwind than a stolen card, because the bill for someone else's treatment arrives months later under your name. The second version is quieter: a real service that keeps the records after the appeal closes, then sells the aggregate to a data broker. The control is the one Fight Health Insurance built in from the start, which is to process the documents on the patient's own device where possible, hold only what the letter needs, and delete on close, and the test for any service is whether it can tell you, in writing, what it keeps and for how long. FLINT Scout runs that check on an agent before the money or the records move, for a penny.

NEXT ISSUE

The freight broker's margin: a truckload moves across the country because a broker phoned a dozen carriers, negotiated a rate, and kept the difference. Two builds now make those calls with a voice agent and book thousands of loads a month without a person on the line, and the difference they keep is the whole fight.