Sooner or later it happens. You open a case and there’s something new on the screen. A score. A summary. Maybe a little sparkle icon. Nobody told you about it, or they did, and the email is buried under forty others.
And the question in your head isn’t really “what is AI in utilization management?” It’s: what is this thing, and should I trust it?
This guide answers both. I work in prior authorization on a Medicare and Medi-Cal team, and I built a continuing education course on this subject, so I’ll keep it practical. No robots in lab coats, I promise.
A short definition
AI in utilization management (UM) is software that reads clinical and administrative data and produces a recommendation, a score, a summary, or a completed task that a nurse or physician used to produce by hand. In plain words: the work that used to take you twenty minutes and three logins.
Utilization management itself is the process a health plan or hospital uses to confirm that care is medically necessary and delivered in the right setting.
That short definition covers three very different technologies. Knowing which one is in your tool matters more than knowing the vendor’s name.
The three kinds of AI you will meet in UM
Think of them as three new coworkers with very different personalities.
Predictive AI: the one with a number for everything
A model trained on past cases estimates a likelihood: that a request meets criteria, that a patient will meet inpatient level of care, that a denial will be overturned on appeal. You see it as a number or a flag.
In a nurse-led project at Yale New Haven Health, UM nurses used an AI-generated Care Level Score to guide their conversations with providers about where patients belonged. The monthly average observation discharge rate fell from 16.69% before implementation to 12.75% after (Tuccio et al., 2025). Notice who did the talking. The score did not make the decision. It gave the nurse a sharper opening. If status decisions are new to you, start with my guide to inpatient vs. observation status.
Generative AI: the fast writer who never sounds unsure
A large language model drafts a clinical summary, a denial letter, or an appeal from the chart. It needs a person to prompt it and a person to check it, because it writes with the same confidence whether it’s right or wrong.
This is the technology most people mean when they say “AI” in 2026, and it has been tested surprisingly little on real patients. A JAMA systematic review of 519 studies published between January 2022 and February 2024 found that only 5% evaluated these models using real patient care data (Bedi et al., 2025). Most were graded on exam questions and made-up cases, not on a real UM queue.
Agentic AI: the new hire who wants to do the whole job by lunch
The newest category chains several steps together toward a goal with limited human prompting: read the intake, pull the record, apply the plan’s criteria (such as MCG or InterQual), then approve the request or route it to a clinician.
Vendors are marketing this now. When someone tells you a tool is “agentic,” your next question is which steps it completes on its own, and at which step a human looks.
What AI in UM is not allowed to do
This is the part nurses most often get wrong, usually because someone at a conference, or in the break room, told them AI is coming for their badge.
California’s SB 1120, the Physicians Make Decisions Act, took effect January 1, 2025. It applies to health plans and disability insurers, including the groups they delegate UM to. Under it, an AI or algorithm tool may not deny, delay, or modify health care services based on medical necessity. Only a licensed physician or a licensed health care professional competent to evaluate the specific clinical issues can make that determination (SB 1120 bill text).
Other states are following. Becker’s Payer Issues counted seven state laws passed in 2026 that limit how insurers use AI in coverage or payment decisions. Washington, Iowa, Colorado, Alabama, and Georgia bar AI from being the sole basis for a denial or adverse decision. Utah adds AI disclosure and independent-judgment requirements, and Illinois targets automated claim downcoding (Emerson, 2026).
At the federal level, the HHS Office of Inspector General’s February 2026 Medicare Advantage compliance guidance is voluntary, but its message is clear: medical necessity determinations must be based on the individual member’s circumstances, and a plan may not rely solely on an algorithm or software that does not account for them (HHS OIG, 2026).
So here’s the honest version. AI in UM can approve, it can prepare, and it can prioritize. In California, it cannot be the one that denies, delays, or modifies care for medical necessity, and a growing number of states are drawing similar lines. A clinician makes that call, and that clinician’s name is on the determination.
How widely is it used?
If it feels like AI showed up everywhere at once, that’s because it pretty much did. Federal data show that 71% of non-federal acute care hospitals used predictive AI integrated with their electronic health record in 2024, up from 66% in 2023 (Chang et al., 2025).
On the payer side, a survey of 93 health insurers by regulators in 16 states found that 84% use AI or machine learning in some capacity (NAIC, 2025a). Among companies selling individual major medical coverage, 71% use, plan to use, or are exploring AI for utilization management (NAIC, 2025b).
The catch is that adoption has moved faster than evidence. The same federal hospital data show that most hospitals evaluate at least some of their predictive models for accuracy, but fewer evaluate all or most of them. That’s why the next section matters.
Four questions to ask about any AI tool in your queue
Before you trust the shiny new box, ask these. They work on vendors, on managers, and on that one coworker who says “well, the algorithm said so.”
Who measured it? The company, a customer, or an independent team? When Mayo Clinic Health System checked a vendor’s diagnosis-related group (DRG) prediction tool against 930 of its own inpatients, the tool’s top-3 accuracy at 24 hours was 71%, compared with the 81.1% originally reported (Muhanga et al., 2025). Even that customer study had a co-author from the vendor. Read the author list. Yes, all of it.
Against what baseline? Is there a before-and-after comparison in a population like yours? A great result in a large academic system may not hold up in a small Medi-Cal plan.
What is missing? If the slide deck shows approval speed, ask for denial rates, appeal overturn rates, and how often nurses override the tool. What a vendor leaves out of the demo is often the most useful part.
Where does the human enter? Ask to see the exact screen a nurse sees when the tool does not approve, and what the tool has already written there. A summary you read first can shape your judgment before you open the chart. The American Nurses Association’s 2026 AI think tank listed erosion of professional judgment through overreliance on AI outputs as one of the profession’s top risks (ANA, 2026).
Will AI replace UM nurses?
No. But it is taking the parts of the job nobody will miss: the requests that clearly meet criteria, the chart hunting, the re-keying, typing the same member ID for the third time.
What it leaves behind is harder: borderline reviews, specific denial rationale, peer-to-peer preparation, and the governance work of testing and monitoring the tools themselves. The ANA’s think tank named advancing AI literacy and competence as one of five near-term actions for the profession for exactly this reason.
The nurses who will be most valuable are the ones who can tell when the tool is wrong, say so clearly, and show their reasoning. AI can draft the note. It can’t defend it on a peer-to-peer call.
Go deeper
If you are considering UM as a specialty, start with the free UM Career Starter Kit.
If you already work in UM, prior authorization, or case management, my course The Role of Artificial Intelligence in Utilization Management – Enhancing Healthcare Decision-Making covers everything in this article in depth: 3 BRN-approved contact hours (CEP #18046), intermediate level, $129, rebuilt in September 2026 from 2025 and 2026 sources, with every reference list downloadable.
Want the full path? The UM Mastery Bundle includes this course plus my foundational and flagship UM courses.
References
American Nurses Association. (2026, May 5). American Nurses Association calls for nurse-led guardrails on artificial intelligence in healthcare [Press release]. https://www.nursingworld.org/news/news-releases/2026-news-releases/american-nurses-association-calls-for-nurse-led-guardrails-on-artificial-intelligence-in-healthcare/
Bedi, S., Liu, Y., Orr-Ewing, L., et al. (2025). Testing and evaluation of health care applications of large language models: A systematic review. JAMA, 333(4), 319–328. https://doi.org/10.1001/jama.2024.21700
California Legislature. (2024). SB-1120 Health care coverage: utilization review (2023–2024 Reg. Sess.). https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202320240SB1120
Chang, W., Owusu-Mensah, P., Everson, J., & Richwine, C. (2025). Hospital trends in the use, evaluation, and governance of predictive AI, 2023–2024 (Data Brief No. 80). Office of the Assistant Secretary for Technology Policy. https://healthit.gov/data/data-briefs/hospital-trends-use-evaluation-and-governance-predictive-ai-2023-2024/
Emerson, J. (2026, July 28). 7 AI health insurance state laws passed in 2026. Becker’s Payer Issues. https://www.beckerspayer.com/policy-updates/7-ai-health-insurance-state-laws-passed-in-2026/
Muhanga, A., King, J., Mueller, J. L., Sanger, J. M., Bartlett, B. N., Manz, J. W., Spiten, M. J., Overgaard, S. M., & Wang, H. (2025). 355 Validation of an artificial intelligence algorithm for predicting diagnosis-related groups in a community health system. Journal of Clinical and Translational Science, 9(Suppl. 1), 109. https://doi.org/10.1017/cts.2024.981 (Author list corrected 2026: https://doi.org/10.1017/cts.2026.10689)
National Association of Insurance Commissioners. (2025a, May 20). NAIC survey reveals majority of health insurers embrace AI [Press release]. https://content.naic.org/article/naic-survey-reveals-majority-health-insurers-embrace-ai
National Association of Insurance Commissioners. (2025b). Health insurance artificial intelligence/machine learning survey results. https://content.naic.org/sites/default/files/inline-files/NAIC%20AI%20Health%20Survey%20Report%20.pdf
Tuccio, L., Catapano, T., Elwell, J., Dupont, N., Sines, E., & Pisanelli, F., Jr. (2025). Leveraging artificial intelligence to improve clinical appropriateness of inpatient designation in a utilization management setting. Journal of Doctoral Nursing Practice. Advance online publication. https://pubmed.ncbi.nlm.nih.gov/41022622/
U.S. Department of Health and Human Services, Office of Inspector General. (2026, February 3). Medicare Advantage industry segment-specific compliance program guidance. https://oig.hhs.gov/compliance/ma-icpg/