AI is breaking into one of healthcare’s most frustrating bottlenecks — and somehow making things worse financially. That’s not a bug. That’s the whole story. If you thought faster processing meant cheaper healthcare, you haven’t been paying attention.
A new report from the Peterson Health Technology Institute has dropped a finding that should make every hospital CFO sweat: AI tools are accelerating prior authorizations and medical coding, but they’re simultaneously driving up costs for health systems. Faster, yes. Cheaper, no. Welcome to machine learning in the real world — where efficiency gains and cost savings are not the same thing, and vendors know exactly which one they’re actually selling you.
What’s Actually Happening Here
Prior authorization is healthcare’s most hated bureaucratic ritual. Doctors hate it. Patients hate it. Insurance companies pretend to hate it while quietly depending on it to delay and deny care. It’s a system designed to slow things down, and for years that slowdown has been the point.
Now AI is stepping in with automation tools that can process prior auth requests faster than any human team. Medical coding — the process of translating clinical notes into billing codes — is getting the same treatment. Machines reading charts, spitting out codes, submitting claims. All of it quicker. All of it slicker.
Sounds great. It isn’t.
The PHTI report found that while these tools do what they say on the tin — speed things up — the downstream financial impact on health systems is negative. Costs are going up, not down. And that gap between what AI promises and what it actually delivers to the bottom line deserves a much louder conversation than it’s getting.
The Speed Trap
Faster Isn’t Free
Here’s where things get interesting. AI tools don’t come free. Health systems are paying real money — licensing fees, implementation costs, integration headaches, staff retraining — to deploy these platforms. When the ROI doesn’t materialize, those upfront investments become millstones.
Vendors have been incredibly good at selling speed. Fewer denials! Faster approvals! Less administrative drag! What the pitch decks bury in footnotes is the total cost of ownership. By the time a health system realizes the numbers don’t add up, they’re already locked into multi-year contracts.
More Approvals, More Spending
There’s another uncomfortable dynamic at play. When prior auth gets faster and approvals flow more freely, more procedures get approved. More procedures mean more spending. Insurance companies process denials faster too — meaning the whole machine runs hotter in both directions. AI didn’t fix the system. It turbocharged it.
It’s a bit like how Amazon’s Project Houdini approach to rapid AI infrastructure promises speed at scale — but speed at scale still requires serious capital. Moving fast costs money. Always has.
The Hot Take
The healthcare AI industry is running the oldest con in tech: sell the efficiency dream, collect the subscription fee, and let the buyer figure out why their margins are bleeding. Health systems are not technology companies. They don’t have armies of engineers to audit vendor claims or optimize integrations. They trust the pitch. They sign the contract. They eat the cost. The PHTI report is essentially documenting an industry-wide fleecing in slow motion, and the silence from AI vendors in response is deafening. If your product speeds things up but makes your customer poorer, you haven’t solved a problem — you’ve created a more expensive one.
What Gets Lost in the Noise
Behind the spreadsheets and cost projections, there are patients. People waiting on approvals for medications, procedures, scans. If AI genuinely speeds up those approvals, that’s real human value — regardless of what it does to a system’s operating costs. That part should not get lost.
But there’s a critical difference between AI that helps patients and AI that helps vendors close their quarterly numbers. Right now, the evidence suggests we’re getting more of the latter than the industry would like to admit. And it connects to a broader pattern of tech products that sound transformative on stage but land differently in practice — whether that’s the medical world grappling with what “breakthrough” really means or AI systems revealing uncomfortable truths about their own limitations and biases.
Where This Goes Next
Health systems aren’t going to stop buying AI tools. The administrative burden is real, the labor market for coders and auth specialists is tight, and the pressure to automate is only growing. What needs to change is the rigor around evaluation. PHTI’s work is exactly the kind of independent scrutiny this market needs more of — not vendor-sponsored case studies, not cherry-picked pilot results, but hard data on whether these tools actually move the financial needle in the right direction.
The machine learning hype cycle in healthcare is not slowing down. But the evidence is starting to catch up with the promises. Health systems that treat every AI vendor pitch with the same skepticism they’d apply to a pharmaceutical rep are going to be in a much stronger position than those chasing speed for speed’s sake. The question isn’t whether AI can process a prior auth faster. It’s whether anyone is actually getting healthier — or wealthier — as a result.