Did you know that four in five physicians now have AI implemented in their practice? They actually do. The exact number is 81%, which was 38% three years ago. However, over the same period, providers reporting denial rates above 5% nearly doubled, as per HFMA and Guidehouse research.
These two numbers shouldn’t coexist, yet they do.
I run MedCare MSO, and we built AI for healthcare, this paradox should trouble me less than it does. The early years were about adoption and assumptions: whether to embrace this change. Later on, when it was implemented, physicians thought value would follow.
Adoption happened, the value didn’t follow and I think I know why.
What Adoption Actually Brought Us
I will be honest: AI has genuinely brought results, and the wins are real. They are just concentrated on one side of the ledger.
Better Clinical Experience
Ambient documentation is the clearest success story in healthcare AI so far. A JAMA Network Open research study found that clinicians at Mass General Brigham using ambient AI scribe resulted in a 21% reduction in burnout in under a quarter. I won’t diminish that. Burnout is a workforce crisis and any technology that gives back physicians their time is worth acknowledging.
Denials Resulting in Revenue Loss
The business case is a little different. 74% of providers report an increase in prior authorization delays, according to the 2026 HFMA/Guidehouse Revenue Cycle Management Trends survey. Yet, the 2025 Bain & Company survey reported that only about one in five organizations applies AI to denial management, even with such high denials.
The real mistake, in my opinion, is that practices hurry in adoption and make the clinical experience better but lag when it comes to fixing the revenue problems. The physician’s burnout was taken care of and a large part of the financial aspect was left untouched.
Where Practices Go Wrong With AI
So where exactly does the gap exist between adoption and value? In my opinion, it’s not something that occurs due to technology failure, it’s the usual purchasing pattern. Most healthcare organizations buy AI the way they buy any other product, as a feature add-on. The workflow stays the same as they were, AI just solves one aspect of it instead of fixing the larger part of the billing cycle.
Why Denials Occur Even After AI Implementation
Let’s say a practice implemented an AI scribe and it does its job well. The documentation is great but the existing coding process, manual charge-capture, and same old school claim submission won’t add any value. There is a need for proper workflows at every stage of the revenue cycle. If not, then there’s no benefit of saving twenty minutes as the claim will still come back denied.
The Problem With Payer’s AI
Payers are also well equipped with AI to review submitted claims, and this conversation is avoided most of the time. The payer’s AI-driven adjudication engine now rejects claims with even minor discrepancies within a matter of seconds. What’s problematic is that appealing to the pace of the payer’s AI rejection isn’t easy. For a practice whose business model is just a one-point solution, it isn’t going to compete iņ such an environment.
Three Things That Can Get Real Value From AI
Building and deploying AI systems for our clients has taught me the difference between what works and what doesn’t. To get real value out of AI, you need to make these three decisions.
1. Fix the Complete Billing Cycle, Not One Step
Documentation, coding, claim submission, denial prediction, and payment posting are not separate problems, and fixing one won’t fix the rest. You need a proper AI ecosystem like ours to be put in place in order to make sure no step is creating problems for the rest.
Our AI coder learns from denial patterns and our claim scrubbing workflows learn from payer behavior and make the following steps smarter and more efficient. If I were to put it simply, the unit of transformation is the workflow, not the task so make sure everything is connected and gives value to others.
2. Judge AI by Financial Outcomes
Hours saved is a great metric and is exactly what you need. Time is money, there’s no denying that. However, the numbers that actually matter are cost to collect, first-pass claim rate, days in A/R, and denial overturn rate.
My advice to any practice evaluation AI is that if the automated solution that you implemented can’t move these numbers within two quarters to betterment, it’s not worth it. You need real results in order to make your investment worthwhile.
3. Your Team Decides If AI Succeeds or Not
The AMA’s 2026 survey found that 88% of physicians now worry about losing their skills to AI, and 85% want a voice in how it gets adopted. These are significant numbers and hold weight. What this means is that if technology is forced on a team, it will quietly diminish their will. However, if technology is built with a team spirit, it will get adopted efficiently and will produce great results.
The goal of AI in the revenue cycle isn’t to remove your team, it is to move them up. Your team will be able to prevent claims denials before they happen instead of chasing their statuses every day. This will only happen when your staff trust the system and feel valued. So get them on the same boat and generate actual value from AI.
Ending Note
The paradox in this article has a simple resolution. Physicians adopted AI, but practices failed to transform workflows around it. Until that changes, denial rates won’t reduce and will keep on climbing no matter how good the AI tool is.
The next five years won’t set you ahead just because you have the most AI tools. Successful practices will be those that connect their entire revenue cycle around AI, judge by its financial outcomes, and bring the entire team along for the change.
This is the standard we, at MedCare MSO, hold ourselves to. And I would encourage every practice leader reading this to hold their AI partners, including us, to the same one.
