A $23 million Department of Justice False Claim Act settlement was traced back to a single automated coding rule in 2025. Yes, an automated mistake with no human oversight. What happened was that the algorithm upcoded thousands of emergency department (ED) visits and CMS flagged the outlier.
This settlement has changed the industry’s perspective towards autonomous AI coding. AI adoption is at its peak right now in healthcare but factors like these are making practices reconsider or be careful with the transition. One important step is HITL (human-in-the-loop) AI which adds a layer of human oversight to autonomous coding.
What Does Human-in-the-Loop AI Actually Mean?
For starters, this is a workflow architecture that lets AI handle the repetitive, high-volume, and pattern-based coding tasks and leaves an RCM expert to do defined points. These points include low confidence scores, billing steps where there’s high compliance risk, and other complex encounters. Just to be clear, this isn’t like AI-assisted coding where a human was in charge of the complete oversight. This is a division of responsibilities among autonomous systems and humans.
HITL AI in RCM Progress Till Now
The very first version of HITL in RCM was actually computer assisted coding (CAC). This emerged after 2010 during the ICD-10 transition and the architecture includes human review on every single claim. You can say it was HITL by necessity, not design as the AI wasn’t trusted enough to act on its own autonomously on anything.
Then came the era following 2019, where the modern HITL architecture began shaping and confidence scoring was introduced. The modern, current era is the industry wide adoption is happening following 2024 and is going fast and steady.
How Does Human in the Loop Work?
The process is very straightforward, every AI code suggestion carries a confidence score and claims scoring above the threshold automatically pass through. Claims that fall below it are routed to an RCM expert who reviews AI’s suggestion, the rationale, and the supporting documentation. After that, the expert approves, edits, or escalates.
Every correction that the expert does eventually feeds back into the architecture and improves its accuracy over time. In result, the system gets smarter with every human decision made and reduces the chances of mistakes. Last but not least, a complete audit trail logs every decision for better compliance and payer defensibility.
Why the Industry Has No Choice But to Adopt it
The False Claims Act does not distinguish between intentional fraud and algorithmic error. It’s either way a liability regardless of who and what generated the false claim. Payers are moving decisively now as Humana and Cigna now require human coder attestation on AI-generated codes as an official, contractual obligation.
CMS’s IPPS proposed rule for FY2026 is expected to formalize autonomous coding oversight requirements as well. Not to mention, the DOJ’s 2025 enforcement posture has expanded to cover AI-powered billing errors. Put it simply, the compliance window is now closing fast for running autonomous coding without human oversight.
The Results HITL AI has Produced Across the Revenue Cycle
As for the financial case for HITL, it is as promising as the compliance case. The organizations that have implemented HITL AI are getting better, measurable results across every major RCM metric. But do know that this is not a marginal improvement, they are an operational transformation.
The numbers we have yet from implementations are reflected from Auburn Community Hospital case study.
Up to 50% reduction in discharged-not-final-billed cases, which were a major issue back in the day. 40% improvements in coder’s productivity and 4.6% increase in case mix index which delivered $1M+ ROI more than 10x their initial investment.
Conclusion
This was everything there is to know about Human-in-the-Loop AI for coding and RCM. With my more than a decade-long experience, I can vouch for it and think that this is going to be the industry standard for autonomous coding. The coder’s fear is no longer there as this technology answers the burning question of whether “AI will replace coders or not.”
So for practices that haven’t yet implemented this architecture in the revenue cycle, I strongly suggest that you do. Every coder here at MedCare MSO does the same and that’s why our medical billing services are always compliant and sure to get maximum reimbursements.
