Prior authorization is one of the most obvious AI automation targets in healthcare — high volume, rule-based on paper, and a genuine source of friction for patients, providers, and payers alike. It’s also one of the categories where AI adoption is real but still modest: AI-prepopulated technical appeals sit at 21% adoption, AI-prepopulated clinical appeals at 19%, and AI denial prediction — a related but distinct capability — at 25%, with the slowest year-over-year growth of any category tracked (4%). That’s a telling pattern: the categories closest to actual payment and clinical determination are adopting AI the most cautiously, even though they’re often cited as the highest-value automation target.

Why prior authorization resists automation more than it looks like it should

On paper, prior authorization looks like a rules engine problem: a set of clinical criteria, a set of documentation requirements, a determination. In practice, the rules vary by payer, change frequently, and require judgment calls at the margins that a purely rules-based system handles poorly — which is exactly why fully automating the determination itself has moved more slowly than automating the administrative work around it, like drafting the appeal or prepopulating the documentation.

That distinction is the useful one: automating the preparation of a prior authorization request or appeal — pulling the right clinical documentation, formatting it to the payer’s requirements, flagging what’s missing — is a fundamentally different, more tractable problem than automating the determination itself. The adoption numbers reflect that: prepopulation and appeals assistance are growing, while full automated determination remains rare and, appropriately, cautious.

Where the real near-term value is

The organizations getting measurable value from AI in this space right now are mostly automating the parts of the process that are genuinely rules-based and low-risk if imperfect — assembling documentation, checking completeness against payer-specific requirements, flagging likely denials before submission so staff can address gaps proactively — rather than trying to automate the clinical judgment call itself. That’s a narrower ambition than “automate prior authorization,” but it’s the version that’s actually producing adoption growth right now, and it compounds: faster, more complete initial submissions reduce the downstream appeals volume, which is where a lot of the administrative cost actually sits.

What to automate first, and what to leave staffed

A useful test for any prior authorization automation initiative: does this step involve assembling and formatting information according to known, stable rules, or does it involve a judgment call about whether a case meets criteria that are genuinely ambiguous or payer-specific and shifting? The first category is where AI is delivering real, growing value today. The second is where full automation is still catching up to the technology’s promise — and treating the two as the same problem is the most common reason prior authorization automation projects underdeliver against their initial pitch.

The practical starting point

Rather than pursuing full prior authorization automation as a single initiative, the more productive path is mapping the specific process into its rules-based and judgment-based components, and automating the first category aggressively while keeping the second staffed and simply better-supported by AI-assembled documentation. That’s a less dramatic pitch than “AI-automated prior authorization,” but it’s the version that’s actually working at the adoption rates the data shows.

SPAR builds revenue cycle and prior authorization automation around this distinction — automating what’s genuinely rules-based, augmenting what still requires judgment — because conflating the two is where most automation initiatives in this space stall. If your prior authorization automation effort has plateaued, that’s often exactly where the gap is.