Here’s a number that should bother you more than it probably does: 1.2 seconds. That’s the average time Cigna doctors reportedly spent reviewing each of the 300,000-plus payment requests flagged by the company’s PXDX system over a two-month stretch, before signing off on the denial, according to reporting cited in a federal lawsuit against the company. Not a typo. One point two seconds to say no to someone’s medical bill.
Whichever way that particular lawsuit goes, the thing it points to isn’t really in dispute: payers are running revenue cycle management through automated systems fast enough that manual review is barely a formality on their end anymore. Meanwhile, a lot of provider-side billing teams are still fighting denials one spreadsheet row at a time. That gap, more than any vendor’s AI pitch, is the real reason to care about this topic right now.
So this isn’t going to be a breezy list of ways AI will change healthcare forever. It’s the specific stuff worth actually knowing: where AI is earning its keep on the provider side, a 2026 federal rule already changing how fast payers have to respond, an honest answer on whether it’s coming for billing jobs, and the one mistake insurers are currently being sued over that you really don’t want to repeat.
The scale of this is easier to see in the numbers than in any press release. A U.S. Senate Permanent Subcommittee on Investigations report found UnitedHealthcare’s denial rate for post-acute care more than doubled, from 10.9% in 2020 to 22.7% in 2022, the exact window the company was rolling out automation across its review process. A separate class action, Estate of Lokken v. UnitedHealth Group, goes further, alleging the company’s AI tool, nH Predict, had an error rate as high as 90%, with roughly nine out of ten appealed denials eventually getting overturned. A federal judge let part of that case move forward in 2025.
Cigna’s version of this story runs through PXDX, the system behind that 1.2-second figure above. The company disputes the AI label entirely, describing PXDX as claims-matching technology it’s used for over a decade, not machine learning. A California judge allowed part of that case to proceed in February 2025 too. Both companies deny wrongdoing, both cases are still working through federal court, and neither outcome changes the pattern underneath them: claims review at scale is automated now, on the payer side, whatever anyone wants to call it.
I’d stop short of calling AI itself the villain here. The actual problem is asymmetry. One side of every claim transaction is running at machine speed. The other side, too often, is still opening PDFs by hand.
Most “AI in RCM” content out there predates this next part entirely, which is exactly why it’s worth including. CMS finalized a rule, quietly labeled CMS-0057-F, that’s reshaping how a huge slice of payers have to handle prior authorization. Not on some future roadmap. This year.
Notice what’s missing from that list: commercial and employer-sponsored plans. If a good chunk of your patients carry employer coverage, this rule barely touches your day-to-day. If you’re heavy Medicare Advantage or Medicaid, it changes a lot, faster responses, mandatory reasons, an API on the way. Worth actually mapping your payer mix against that list before assuming either way.
Short answer: no, not wholesale, and not soon. Longer answer, and the more useful one: AI is absorbing the repetitive, pattern-matching parts of the job, initial code suggestions, eligibility lookups, flagging likely denials, while the parts that require judgment, resolving an ambiguous documentation gap, deciding whether an appeal is worth pursuing, catching something a model got wrong, still need a person standing behind them.
The UnitedHealthcare and Cigna situations above are actually the clearest argument for why the human stays in the loop instead of getting removed from it. A tool with a reported 90% reversal rate on appeal wasn’t replacing human judgment successfully. It was making expensive, high-volume mistakes at a speed nobody caught in time. The realistic shift isn’t fewer coders. It’s coders spending less time on repetitive entry and more time on the exceptions AI flags but can’t resolve on its own.
There’s a specific trap worth naming directly here: adopting AI in RCM without keeping a human reviewing its output is exactly the mistake currently playing out in federal court against two of the largest insurers in the country. It’s tempting to buy an AI coding or denial-prediction tool and treat its output as final, especially when staffing is already stretched thin. The lawsuits above are a live example of what happens when nobody’s checking the model’s work at scale.
The practical version of this: use AI to triage and prioritize, not to make the final call unsupervised, at least not yet. A denial-prediction model that’s wrong 10% of the time and gets human review before submission is a genuine improvement over the status quo. The same model running unsupervised at scale is a liability with your name on it instead of a payer’s.
None of this is a small IT project you casually bolt onto an already-stretched billing team. Before adopting any AI tool in your revenue cycle, regardless of who’s selling it, a few plain questions are worth asking out loud, and worth insisting on real answers to. Does an actual person review the output before a claim goes out, or does the tool just submit on its own? Can the vendor show you their real error rate, not the accuracy number from the sales deck? Is patient data handled in a way that’s genuinely HIPAA-compliant, not just described that way in the brochure? And does it plug into your existing EHR and clearinghouse, or does it just become a second system your staff now has to babysit separately?
If a vendor can’t answer those cleanly, that’s the answer.
AI isn’t the problem. Unsupervised AI is.
Used correctly, AI can spot denial risks before they cost you, accelerate coding and eligibility checks, and help your team focus on the claims most worth recovering. But technology alone doesn’t protect your revenue. The right people, processes, and oversight do.
That’s the Scintillate RCM difference.We combine intelligent automation with experienced RCM professionals who know when to trust the technology—and when to question it. So you get the speed of AI without handing your revenue cycle over to a black box.
Less chasing. Fewer preventable denials. More revenue recovered. And you don’t have to figure out what to automate, what to keep human, or where your revenue is slipping away. We’ll find it with you.
Your revenue worked hard to get here. Make sure it makes it all the way to your bottom line.
Talk to Scintillate RCM Healthcare and discover where your revenue cycle can work smarter, faster, and harder for your practice.
Not wholesale, and not soon. AI is absorbing the repetitive parts of the job, initial code suggestions, eligibility checks, flagging likely denials, while judgment calls, resolving documentation gaps, deciding whether to appeal, still need a person.
The most common uses today are predictive denial flagging before a claim is submitted, coding assistance that suggests codes for a human coder to confirm, real-time insurance eligibility verification, and prioritizing which aged denials are worth appealing.
Insurers including UnitedHealthcare and Cigna face active lawsuits alleging their AI and automated review systems led to increased or improperly reviewed claim denials. Some of these cases are still working through federal court, and the companies dispute elements of the claims against them.
For straightforward eligibility checks, real-time AI verification is generally reliable and faster than a manual portal lookup. It doesn’t remove the need for a human to handle edge cases, like a plan change mid-cycle or a coverage dispute, but it catches the routine lapsed-coverage problem before it becomes a denial.
CMS-0057-F now requires Medicare Advantage, Medicaid, CHIP, and ACA exchange plans to respond to prior authorization requests faster (72 hours expedited, 7 days standard) and provide a specific denial reason. Commercial and employer-sponsored plans outside those categories aren’t covered by the rule.
Free RCM Health Check. Receive a custom benchmark report comparing your performance against industry standards in 2 days.
Scintillate Healthcare LLC, Georgetown, Texas.
Cash Accelerated Care Elevated
