In 2024 we said effective AI in healthcare RCM needs a human on the final call. Payer AI has since made this an essential condition of the work.
By: Ritesh Ramesh, CEO, MDaudit
Throughout 2024 I wrote that effective AI in healthcare revenue cycle management (RCM) requires a human in the loop. Not as a courtesy to the people the software might displace, but as a core operating condition. The case ran first on Unite.AI in January, returned through the year, and closed the fall in Becker’s Hospital Review and on RACmonitor. Autonomous AI in RCM is a pipe dream. Vendors who ship models without understanding the underlying RCM workflows produce noise, and three things will always need a person: a sound data foundation, continuous training against a moving regulatory target, and governance that keeps a human on the final determination.
Two years later, that is no longer a contrarian position. It is how the work needs to be done. What changed is the market around it. Payers have deployed AI against claims at a volume no manual program can answer one letter at a time. Authorization denials have given way to line-item underpayments, downcoded claims, and disputes over DRG assignment. Our analysis of 2026 payer audit activity found coding errors behind nearly seven in ten completed denials, with payer audit volume and dollars at risk up 30 percent year over year. It’s tempting to answer payer-side AI with autonomous AI of our own. That reads the last two years backwards.
The failure mode has not been retired. In 2024 I described a customer whose so-called autonomous coding system was miscoding drug units. The error had already cost $25 million, and it was caught because a human was still in the loop. Models are faster now. The costs of uncaught errors are not smaller. Poorly governed AI still misses revenue, still overcodes into a compliance problem, and still labels the wrong physician or facility as the risk. Speed without a place for judgment scales all three.
“Humans in the loop” gets used now as a disclaimer, the sentence a vendor adds so the demo feels safe. What it should be instead is an applied philosophy of product architecture. The model finds the needle across millions of claims. A person makes the determination that has to hold up when a payer, regulator, or physician asks why. Skip the data foundation and the model hallucinates. Skip continuous training and the system is enforcing yesterday’s rules. Take the human off the final call and the organization owns an output no one reviewed.
We had already been building to this before the 2024 pieces. AI Assist and SmartScan.ai shipped in 2023. Auditor Assist, added this year to our continuous risk monitoring suite, is the latest expression of the same standard, not a change in direction. We call that standard Meaningful AI. Before anything ships, the question is whether it changes the outcome: a clear line to ROI, less operational friction, and a human who keeps the final say.
AI Assist is how a revenue integrity team talks to its own data. A compliance lead or an auditor asks a question in plain language and gets a data-rich, contextual answer from claims, denials, and audit history, without waiting on a technical interpreter. SmartScan.ai is how the external audit stops dying in intake. It reads the payer request letter, extracts the account data, and puts it into the workflow, so the hour goes to the response rather than to rekeying. In its first year, SmartScan.ai helped clients respond to more than $100 million in audit requests. Payer Audit Workflow, which it sits inside, helped customers retain more than $375 million in 2025. Auditor Assist is how AI joins the audit without taking it over. It reads the medical record against the coded claim, recommends where coding integrity is at risk, and learns from the auditor’s decision. It does not make the determination. Every output is sourced and traceable, and the auditor decides what happens next.
As coding automation spreads on both sides of the claim, the question stops being whether a model touched the code. It becomes whether anyone can show their work. An output that cannot be traced cannot be defended. In 2024 I wrote that AI will transform healthcare RCM only if organizations augment the technology with the workforce that knows the work. The window in which that could be dismissed as a hedge has closed, because payer AI closed it. We have been saying that from the start. The platform is how we keep saying it.
Contact us today to learn more about how the MDaudit platform helps health systems minimize billing compliance risks and maximize revenue.
