HSS Claims AI Cut Appeals from 45 Minutes to Five, Lifted Success Rate to 100%

Hospital for Special Surgery had a problem it could not solve with people. Its insurance appeals were taking 45 minutes each, succeeding only two-thirds of the time, and required a team of staff plus a third-party contractor to handle the volume. Then it plugged in an AI agent.

Nine months later, HSS handles all claims in-house. The AI processes 1,100 claims per month. Appeals that once took 45 minutes take five. The success rate hit 100 percent.

That is the claim. And if it holds up, the $800 billion medical billing industry should be nervous.

The vendor lock-in that agentic AI just broke

Revenue cycle management has been a growth business for decades precisely because the work is complex. Claims get denied. Appeals need to be filed. Payers have armies of reviewers whose job is to find reasons not to pay. Hospitals, already stretched on clinical staff, have historically found it cheaper to outsource this work than to build the internal infrastructure to fight for every dollar.

That calculus is what Ema, a seven-year-old agentic AI startup backed by Accel and Section 32, is now trying to dismantle. Its pitch to hospitals: we will not just automate your billing tasks, we will make your third-party contractor redundant.

HSS is its most prominent reference customer. The New York hospital ranks number one in orthopedics by U.S. News for 16 consecutive years. It announced a three-year engagement with Ema in January 2026. The partnership covers patient scheduling and triage today and, according to the announcement, will expand into HR and IT support over the life of the contract. But it is the claims processing results, surfacing only now in sponsored content published by MIT Technology Review, that are the more consequential data point.

The mechanism matters. HSS did not just automate a step in the existing workflow — it eliminated the contractor layer entirely. The hospital now does in-house what it previously paid a specialized vendor to do. That is workflow-stack collapse, not workflow optimization. And if it generalizes, it is an existential threat to every company whose business model depends on hospitals lacking the infrastructure to handle complex claims processing themselves.

The canary in the billing mine

Waystar, Change Healthcare, and R1 RCM collectively process hundreds of billions of dollars in claims annually. Their growth thesis rests on a simple premise: the complexity of medical billing is growing faster than hospitals ability to manage it internally, and only specialized vendors have the scale and expertise to close the gap.

Agentic AI, if the HSS results hold, attacks exactly that premise. The complexity does not disappear — but it becomes manageable without outsourcing. The KPMG research finding that 68 percent of providers have already adopted AI agents suggests the industry is already moving in this direction whether the RCM incumbents acknowledge it or not.

None of the major RCM players has disclosed meaningful agentic AI disruption risk in their investor communications. The market cap trajectory of publicly traded RCM companies has not yet reflected any concern about a technology-driven disintermediation. That is either because the threat is not real, or because it has not surfaced in enough hospital deployments to show up in the numbers yet.

HSS is one data point. It is a premium specialty hospital with unusually high claims volume and unusually motivated administrators — not a representative community hospital. The metrics are self-reported by HSS, not independently audited. The nine-month evaluation window is short for claims processing ROI, where seasonality and payer mix shifts can distort results over a year.

The sponsorship problem

The MIT Technology Review article that surfaced these results was produced by Insights, the publications custom content arm, in partnership with Ema. The sponsorship is disclosed in the pieces own footer — but the disclosure does not change what it is: a vendor-funded case study presented as editorial content.

Every metric in the article comes from HSS, and every metric benefits Emas pitch. The 65-to-100 percent appeal success rate improvement is a powerful number. It is also not a number anyone outside HSS has verified. The articles own framing — that agentic AI will rehumanize healthcare by giving clinicians back their time — is Emas marketing language, attributed to Dr. Ashis Barad, HSSs chief digital and information officer, who appears in both the press release and the sponsored article.

This does not mean the numbers are false. It means a story built entirely from vendor-funded content with no independent corroboration should be treated as a press release that happens to have bylines.

What would actually prove the thesis

The economics angle here is real and worth pursuing — but it needs more than one hospitals self-reported results to be a story.

The contract between HSS and Ema is not public. A procurement record, a public filings search, or a FOIL request would show whether there are performance guarantees attached to those metrics, which would tell us whether Ema is betting its reputation on the numbers or simply selling software and letting the customer take the credit.

The RCM vendors own disclosures are the other obvious place to look. Waystar filed for an IPO in 2023. Change Healthcare was acquired by UnitedHealth in 2022. R1 RCM is publicly traded. None of their investor presentations, earnings calls, or regulatory filings in the past 18 months show any meaningful discussion of agentic AI as a disruption risk — which is either reassuring or a sign that the risk has not reached the boardroom yet.

And the verification question: has any independent party — an auditor, a health system that evaluated and rejected Ema, a rival vendor — weighed in on whether 100 percent appeal success is plausible? Medical claims appeals involve payer-specific rules, state regulations, and documentation requirements that vary case by case. A 35-point jump in success rate in nine months is either a genuine workflow transformation or a cherry-picked numerator.

The story is not that HSS uses AI agents. Every hospital system in the country is evaluating or deploying them. The story is whether the economics of medical billing are about to change — and whether the companies that built their businesses on the assumption that hospitals could not do this work themselves are about to find out they were wrong.

That story needs a contract. It needs a competitor quote. It needs the RCM industry to respond. Right now it has a vendor-funded article and a hospital that is a satisfied customer.

Run it up the flagpole. But verify first.