The vendor pitch sounds clean. The data tells a different story.
If you run a practice, a health system, or a specialty group, you’ve probably heard the pitch. An AI vendor says it can automate 90% of your prior authorization submissions via payer APIs. Humans only touch the exceptions. Faster, cheaper, fewer errors.
It’s a darn good story. It’s also mostly a story about one narrow slice of the problem.
The Market Was Already “Automated” Years Ago
The first thing worth noting about AI prior authorization is that it is already heavily digital. A large share of prior auth volume, in our own book, roughly 75%, has been submitted through online portals for years. That ratio has been stable since 2023. The shift from phone-based submission to digital submission isn’t coming. It happened a while ago.
What has changed inside that digital workflow is how hard each case is becoming. Denial volumes have risen sharply. Peer-to-peer requirements have increased. Retroactive reviews are more common. Payer criteria tighten year over year. The market isn’t getting easier to automate. It’s getting harder to approve.
Where the Growth Actually Is
Between 2023 and 2025, the total volume of prior authorization work we handled grew by 49%. In the same window, the complex tail of the work, denials and peer-to-peer cases, grew by 307%. Online denials alone grew 623%.
Prior Authorization Growth Divergence
Total volume grew steadily. The hard work grew six times faster. (2023–2025)
+49%
+307%
+623%
In other words, the difficult segment of the market has been expanding more than six times faster than the base.
If submission automation were genuinely simplifying prior authorization, you would expect the opposite. You would expect prior authorization denial volumes to fall as APIs routed more cases through clean eligibility checks. Instead, the opposite is happening at scale. The promise of AI for prior authorization runs straight into a market that is generating harder work faster than it is generating easier wins.
Where Denials Actually Come From
Case review across our book shows that 90 to 95% of denials trace to one of three root causes:
- Client-side documentation deficiencies. Missing chart notes, incomplete clinical justification, unsigned orders, and medical-necessity language that doesn’t meet current payer criteria.
- Member plan limitations. Benefit exclusions, coverage caps, step therapy requirements, exhausted alternatives.
- Non-preferred medications. Formulary restrictions, prior-preferred drugs not tried, clinically appropriate but non-covered choices.
- None of these are problems a submission-layer API can prevent or resolve.
An auto-approval system submits whatever the EMR hands it. If the progress note doesn’t establish medical necessity, the AI doesn’t know, and it can’t fix it. An API doesn’t know the patient’s remaining benefit, step therapy history, or exhausted alternatives. And when a non-preferred medication is involved, no amount of submission speed substitutes for a clinical rationale and a live peer-to-peer with a payer medical director.
The View From the Payer’s Side
There is a second layer to the data that reinforces the point. Since we began tracking the specific reasons payers cite when they deny a case (a practice we started partway through our history, so this sample covers only the most recent period), every single denial in our coded sample, 100% of 1,192 cases, has fallen into clinical-judgment categories.
100% of Denials Cited Clinical Judgment
Every coded denial in our sample fell into a category that requires clinical advocacy to overturn — not faster submission.
This tells us something particular about the denial work itself. The language payers use to deny cases is clinical. The response required to overturn those denials is clinical. Extra documentation, updated chart notes, peer-to-peer discussion with a medical director, and exception requests grounded in clinical rationale. A submission API operates at a different layer entirely. It has nothing to say in a clinical conversation, because it cannot conduct one.
A note on scope: this coded sample does not capture benefit-exclusion or plan-limit denials, which are tracked separately. But among the denials where we have recorded the payer’s reason, 100% required clinical judgment to resolve.
AI in Prior Authorization Doesn’t Eliminate Denials. It Feeds Them.
This is the piece most vendor pitches leave out. When AI in prior authorization submits a case with incomplete medical documentation, a plan limit it didn’t see, or a non-preferred drug, it doesn’t produce a clean outcome. It produces a denial, which then enters the appeal, peer-to-peer, and documentation-chasing workflow.
That’s where the real growth in prior authorization services is happening. “90% automated submission” doesn’t mean “90% fewer problems.” It means more cases moving through submission faster and, in a tightening payer environment, a larger absolute volume of denials landing downstream.
The Work That Genuinely Wins the Case
Every prior authorization that doesn’t approve on the first pass enters a second workflow that most AI vendors don’t acknowledge: interpreting the denial rationale, assembling a targeted response, coordinating peer-to-peer scheduling against payer medical director calendars, and tracking the case through additional rounds if needed.
This is relationship-driven, orchestration-heavy work. It is also one of the capabilities most remote from what a submission API can touch. And it’s where approval rates are actually determined.
What to Ask an AI Prior Authorization “Solution” Provider
If you’re evaluating automation for prior auth, three questions will clarify the pitch quickly:
Is that 90% of total PAs, or 90% of the eligible-for-API subset? Almost always the latter.
What happens to the denied cases and those clawed back during retroactive review? That work doesn’t go away. It just changes hands.
What percentage of denials trace to documentation gaps, plan limits, or formulary issues? Because those are the ones no API can solve.
Key Takeaways
- Roughly 75% of prior authorization submissions in our book have been digital since 2023. The submission layer is not the frontier.
- Total PA volume grew 49% from 2023 to 2025; denials and peer-to-peer work grew 307%; online denials grew 623%.
- 90 to 95% of denials trace to documentation gaps, plan limits, or non-preferred medications. None are solvable at the submission layer.
- Of 1,192 coded denials, 100% cited clinical-judgment reasons. Resolution requires clinical advocacy, not faster submission.
- AI prior authorization tools accelerate the easy part of the workflow while increasing the burden on the more complex parts. The hard part is determining approval rates.
The Bottom Line
Submission automation is real and effective within its defined scope, but addressing only the simplest part of prior authorization shifts greater complexity downstream. The true frontier, where clinical outcomes are decided, lies in the expanding domain that requires human expertise, clinical judgment, and constant advocacy. As the market still evolves, leading in this segment is what will make the difference for both practices and patients.
That’s the segment in which DataMatrix Medical operates.
Nathaniel Smathers is the VP of Client Education and Marketing. He is also a long time contributor of the DataMatrix Medical blog and has a background in healthcare content creation for over a decade. Nathaniel is passionate about exploring the intersections of healthcare, data analysis, and digital innovation.
