AI Is Already Reviewing Your Employees’ Claims. Do You Know What It’s Deciding?

The opportunity, and the fiduciary risk, of AI in claims review and utilization management

Photo by Tiger Lily, Pexels

Somewhere in your health plan’s claims pipeline right now, an algorithm is making a call. It might be flagging a prior authorization request, scoring a treatment against a clinical guideline, or deciding whether a claim needs a second look before it pays out. Your third party administrator probably didn’t ask your permission to use it, and there’s a good chance nobody on your team has ever seen it in action.

That’s not a hypothetical. AI in claims review and utilization management is already embedded across the industry, used by carriers, TPAs, and utilization review vendors to speed up prior authorization, flag billing errors, and support medical necessity decisions. The technology is not the problem. The problem is that most employers sponsoring self-funded plans have no idea it’s there, let alone how it works or who is accountable when it gets something wrong.

You run every other part of your business on strategy and oversight. Your health plan should not be the one place where you hand someone the keys and never ask what they’re doing with them.

The Opportunity: Why AI in Claims Review and Utilization Management Is Growing

Photo by olia danilevich, Pexels

Utilization management exists to make sure health care dollars go toward medically necessary care. Done well, AI can genuinely strengthen that process. It can process claims faster, catch coding and billing errors that a human reviewer might miss, flag inconsistent or fraudulent patterns across large volumes of data, and support clinical reviewers with faster access to relevant medical history so decisions are less delayed for your employees.

For a plan sponsor, that can translate into real value: fewer processing delays, faster prior authorization turnaround for employees who need care, and a claims process that catches waste before it hits your plan’s spend. That’s the pitch, and in the right hands, it holds up.

The catch is that “in the right hands” is doing a lot of work in that sentence, and as the plan sponsor, you are the one legally responsible for what happens in those hands.

None of this is speculative. Utilization management touches every claim that crosses your plan’s desk, and the volume most TPAs process now makes a purely manual review unrealistic. A mid-size self-funded plan can generate thousands of prior authorization requests a year, and a human-only review process at that scale either gets slower or gets less careful. AI-assisted utilization management is one answer to that volume problem, which is exactly why so many vendors have already adopted it, often without asking your plan committee to weigh in first.

The Risk: This Is Your Fiduciary Duty, Not Just Your Vendor’s Product Roadmap

If you sponsor a self-funded ERISA plan, you are a fiduciary. That status does not pause the moment your TPA plugs in an algorithm. Under ERISA, plan fiduciaries are required to ensure a “full and fair review” of every health claim, and to act solely in the interest of plan participants when selecting and monitoring the vendors who administer the plan on your behalf.

That standard, per KFF’s research on AI in prior authorization and claims review, has not yet been clearly defined for AI-driven decisions. Delegating a task to a piece of software does not delegate away that duty.

Regulators are moving too, just not all at the same pace or in the same direction, which is exactly what makes this messy for a national employer. As of early this year, at least 25 states had adopted guidance based on a 2023 model bulletin from the National Association of Insurance Commissioners, which, per that same KFF analysis, sets expectations that AI used anywhere in the insurance life cycle, including claims administration, has to comply with existing fair trade and anti-discrimination laws.

That guidance also requires insurers to be governed by a documented internal oversight program. Some states have gone further and passed their own laws.

California’s SB 1120, sometimes called the Physicians Make Decisions Act, now requires that AI tools used in utilization review cannot autonomously deny, delay, or modify a request for care. A licensed physician or qualified clinician has to make the actual medical necessity determination, and the decision has to be based on the individual member’s clinical circumstances, not just a group dataset.

Photo by Cedric Fauntleroy, Pexels

None of this means AI is off limits for your plan. It means the standard for using it responsibly just got a lot more specific, and the responsibility for meeting that standard sits with you as the plan fiduciary, not only with the vendor who built the tool.

What “Prudent Oversight” Actually Looks Like Here

You don’t need to become an AI or utilization management expert to meet your fiduciary duty. You need a process, and you need to be able to prove you followed it. Courts evaluating ERISA claims tend to focus less on whether every outcome was perfect and more on whether the fiduciary followed a reasonable, documented decision-making process. That’s good news, because it means the bar is about diligence, not omniscience.

Get the answers in writing. Put them in your vendor contracts and your fiduciary file. If your vendor can’t answer these clearly, that’s information too, and it belongs in your risk assessment either way. This is exactly the kind of ongoing vendor monitoring covered under a fiduciary health solutions framework, built so oversight of tools like these does not fall through the cracks between renewal cycles.

Renewal season is the natural checkpoint for this. When you’re already reviewing plan performance, network adequacy, and vendor contracts, add utilization management oversight to that same conversation instead of treating it as a separate project. A vendor that cannot answer the questions above in writing is a vendor whose renewal deserves a harder look, not a rubber stamp.

What Happens If You Wait

Waiting rarely feels like a decision, but it is one. Every renewal cycle you spend without asking your utilization management vendor these questions is a cycle where the fiduciary exposure compounds instead of resolving. If a member ever challenges a denial that ran through an AI-supported utilization management process, the first question a regulator or attorney asks is not whether the outcome was correct. It is whether you, as the plan sponsor, had a documented process for overseeing that tool in the first place.

That documentation does not need to be complicated. A short internal policy, a set of vendor answers on file, and a note in your fiduciary meeting minutes covers most of the exposure. What it cannot be is silence. Silence looks the same whether your utilization management vendor is doing everything right or nothing at all, and that is exactly the position no fiduciary wants to be in when someone finally asks.

This Is Not Just a California Problem

It’s tempting to file this under “state law we don’t have to worry about” if your workforce sits outside California. That would be a mistake. Utilization management vendors build their systems once and deploy them everywhere, so a rule written for one state’s insured plans often reshapes how a vendor’s tool behaves for every plan it touches, including self-funded plans elsewhere.

The NAIC model bulletin guidance already covers a majority of states, and more state legislatures introduce AI-specific utilization management bills every session. Waiting to see if your state passes something similar misses the point. Your vendor’s utilization management practices are already changing in response to the states that have.

The Real Choice in Front of You

Photo by Ivan S, Pexels

You can treat AI in utilization management as background noise your vendor handles, or you can treat it as what it actually is: a decision that affects your employees’ access to care and your standing as a fiduciary. One of those choices is reactive. The other is strategic, and it’s the same posture you already bring to every other part of your business.

Your employees are trusting that when they need care, someone competent is looking at their claim. Make sure that’s still true, whether the first look comes from a person or a model.

This article is for general informational purposes and does not constitute legal advice. Plan sponsors should consult with ERISA counsel regarding their specific fiduciary obligations and vendor oversight practices.