5 Reasons AI Rollouts Stall in Claims and How to Avoid Them

McKinsey’s 2025 global AI survey found that 88% of organizations now use AI in at least one function, but only about a third have gotten past piloting it, and just 39% report any measurable impact on the bottom line.

The reason so few scale isn’t the model, but how it’s rolled out. Here are the five places rollouts get stuck in claims review, and what tends to get them moving again.

Everyone is talking about AI and claims teams know that they need to be doing something with it. They see a demo, watch a chronology get built in seconds instead of hours, and they’re ready to sign up because technology looks impressive.

What often gets skipped is the harder question: What, specifically, is hindering the team today?

Is the bottleneck the organization of an incoming medical record set, drafting a summary overview, or compiling key criteria for the final report? Those are three different problems, and a tool designed to solve one won’t necessarily solve the others.

Without clearly defining the problem first, kicking off a tool can go well and still go nowhere. The software may be useful, but it’s not addressing the issue that’s actually slowing the process down. It’s a quiet failure, if nobody can point to a measurable improvement, there’s no compelling reason to continue the rollout.

The fix: Define the specific problem the AI is meant to solve before evaluating a tool. Ask “if this AI initiative is a success, what does that look like? What changes?” Whatever the answer, determine the metrics for success – fewer hours per file, faster turnaround on decisions, more detail per report.

That gives everyone a clear target and makes it possible to evaluate whether the technology is actually solving the problem it was brought on to solve.

Sometimes the honest answer really is bad timing – a bodily injury claims unit six months into migrating to a new claims management system, doesn’t have the people to run a pilot right now. More commonly, the people needed for sign-off weren’t looped in at the right time, and with each stakeholder juggling their own priorities and projects, the AI initiative slips through the cracks.

Any AI involving PHI usually needs agreement from compliance, legal, and IT. If the champion for the tool doesn’t bring key stakeholders in until the deal is nearly done, the rollout hits a wall and weeks pass waiting on signatures – usually because key players were brought in to approve a decision instead of shaping it from the start.

The fix: Identify every stakeholder who has to say yes before the vendor is chosen, and bring them in early enough that concerns can be addressed without stalling the rollout. Set a decision deadline and work with each stakeholder, assign ownership to tasks, and come prepared with the answers they need, so the process moves quickly instead of stalling out on questions at the end.

A new AI tool is rarely a line item that already exists in this year’s budget. The team has a real problem and found an application that solves it, but showing up with an impressive demo and no numbers is a hard sell to the person who controls spending.

A polished feature set doesn’t answer the question the budget owner is actually asking, which is the return on investment.

The fix: Make the case that the AI is worth funding – pull hours per file, reserve adjustments, appeal rates, denial reversals, or cost per re-opened claim and compare them to what the platform can realistically do. Create the productivity story, then ask the vendor for real numbers from comparable implementations to help your ROI case.

In claims, this is the single biggest driver that delays technology implementation. Compliance can’t be a formality that happens after the fact; it’s part of the sale from the start, because claims decisions carry legal weight and the medical records behind them hold sensitive data. Any tool touching them has to hold up under both HIPAA (U.S.) and PIPEDA (Canada) standards.

Teams that get stuck here let the conversation around security surface too late. The evaluation runs its course, a contract lands on legal’s desk, and the security review hasn’t even started.

There’s a second piece that gets missed: AI can’t be the one making the coverage call. Whatever AI summarizes or extracts needs to trace back to the source record with an audit trail, because a decision that can’t be checked against the file it came from isn’t defensible if it’s ever challenged.

The fix: Start the compliance conversation as soon as a solution is identified, not after a contract lands on legal’s desk. Have the security documentation ready ahead of the review, and surface red flags before they become blockers. Build accountability into the process, ensuring the tool has an audit trail, links to the source record, and the coverage decisions stays with the adjuster.

Rollouts stall when staff see AI as replacing their role or judgement. That framing can turn adoption into a search for the tool’s mistakes rather than focus on where it can help. Fear, skepticism, and defensiveness are common among adjusters who’ve seen “efficiency” initiatives that overpromised, while others share unrealistic expectations of what the tool can do.

The solution typically isn’t intentionally sabotaged, but it can happen when people feel a change was done to them instead of with them.

The fix: Co-design the use case with key users. They know the current process, where it breaks, and early involvement turns skeptics into champions. Be clear on what the AI will handle versus what stays a human call, support the team through the change (not just train them on the software), and build a feedback loop that can help improve the systems outputs.

Start small – with one claim type or team, prove it works, then expand piece by piece instead of flipping the organization upside down trying to do everything at once.

Two-thirds of organizations are stuck exactly where McKinsey’s report puts them: piloting, never scaling. Usually because the rollout got treated as a purchase decision instead of a change in how the work gets done.

The teams that get past the pilot define the problem, set metrics, and bring in the people who’ll use the tool before the contract, not after. If you’re evaluating AI for claims review, success isn’t just about how impressive the technology is – it’s about doing the groundwork before you sign.