For more than a decade, I've worked on the reviewing side of prior authorization — inside managed care, utilization management, and clinical review programs, most recently reviewing somewhere between 50 and 70 PA cases a day. That volume teaches you things a textbook doesn't. It teaches you where the process actually breaks down: not usually in the clinical judgment, but in the surrounding friction — incomplete documentation, criteria scattered across systems, inconsistent reasoning between reviewers, and cases that sit longer than they need to because the information needed to decide them isn't organized in one place.
METHUVIA AI is my attempt to fix that friction without touching the part of the process that actually requires a pharmacist's judgment.
The engine produces recommendations. It never makes the final call. Every coverage determination is made by a licensed pharmacist — the same as it is today, just with better-organized information in front of them.
What Prior Authorization Review Actually Looks Like From the Inside
Most people outside managed care picture PA review as a single yes-or-no judgment. In practice, it's a five-step process repeated dozens of times a day: extract the coverage criteria from the plan's policy, extract the relevant facts from the PA form, extract the clinical facts from the chart note, cross-validate all three against each other, then apply disposition logic to reach approve, deny, pend, or escalate. Every one of those five steps is where errors, inconsistency, and delay creep in — not the final judgment call itself.
What METHUVIA AI Does — and What It Deliberately Doesn't Do
The platform is built around a dual-pass evaluation: the engine independently works through the same five-step process a reviewing pharmacist would, surfaces its reasoning, cites the specific criteria and clinical evidence it used, and flags anything it isn't confident about. The reviewing pharmacist sees that reasoning laid out clearly, checks it against their own read of the case, and makes the actual determination. Every override the pharmacist makes is tracked, and that concordance data feeds back into making the system more useful over time — not more autonomous.
What it doesn't do: it doesn't approve or deny anything on its own, it doesn't remove the pharmacist from the loop, and it isn't positioned as a replacement for clinical judgment. If you've read anything else I've written about AI in pharmacy, this is the same position applied to my own product — AI is genuinely useful for organizing information and surfacing patterns, and genuinely bad at being handed final authority over a patient's care.
Why I Think This Matters
Prior authorization has a real reputation problem, on both sides of the fax machine. Prescribers experience it as an opaque black box. Patients experience it as a delay with no clear explanation. Reviewing pharmacists experience it as a volume problem that pulls their attention away from the cases that actually need careful thought. None of that is solved by moving faster on autopilot. It's solved by making the well-reasoned, well-documented decision easier to reach the first time — and by keeping a licensed pharmacist accountable for every single one of them.
Learn more at methuviaai.com
I'm building out the platform, the underlying engineering spec, and the case for pharmacists, payers, PBMs, and health plans. If you work in managed care, utilization management, or health-plan operations and this resonates, I'd like to hear from you.
Visit methuviaai.com →You can also read more about my clinical background on the About page, or explore my writing on prior authorization and utilization management in the pharmacy articles section.