Revenue Cycle is AI’s Easiest ROI Case, if Your People are Ready

In early October I sat in on a panel of finance leaders at the UT REAL Health AI Symposium in Austin. When the moderator asked where AI is producing economic value today, the first panelist pointed to the revenue cycle: finding denial patterns across payers, checking whether reimbursement matches the contract, and evaluating managed care relationships. Those are hard dollars you can build KPIs around.

Later in the session, the moderator made an observation that stuck with me. Physicians (for the most part) feel more secure about their place in an AI world. Revenue cycle and collections staff feel threatened.

Both make good sense… Revenue cycle is AI’s easiest ROI case, and its workforce is rightfully anxious about it. From the AI literacy work I do with health systems, I’d argue the willingness of the staff to adopt the technology and find opportunities with AI will impact the success of the solutions. If people are educated on AI, re-skilled to better use AI, and given an understanding of where it will fit into the vision for the organization, benefits of AI will be more easily captured.

The tools are live, and the returns are real

Epic introduced Penny, its AI assistant for revenue cycle, at its Users Group Meeting (UGM) in August 2025. A year later, Epic reported 360+ live for its Professional Billing Coding Assistant.

Northeast Georgia Health System’s generative AI appeals tool recovered more than $3 million from a backlog of outpatient claims. One finance leader on the panel described a plan to use administrative AI as a financial flywheel, with revenue cycle returns helping fund AI work where the payoff is less proven.

Still, in Deloitte’s 2026 health care CFO survey, CFOs at only 18% of organizations scaling AI reported that they consistently measure its financial impact. The tools arrived faster than the metrics and capability for adoption.

Significant fear sits in revenue cycle

One panelist noticed a split worth paying attention to with AI ideas bubble up from physicians, but in revenue cycle and finance operations they mostly come from leadership. The moderator added that even inside finance, financial planning and analysis (FP&A) teams welcomed AI because it frees them from building spreadsheet models, while revenue cycle teams were more apprehensive.

That fear makes sense given what we are seeing. In the 2026 Bessemer and Bain ROI scorecard, 73% of providers reporting current or planned AI-related headcount reductions named revenue cycle and medical billing. One panelist expected the revenue cycle workforce to get smaller over time, with the people who learn the tools and fold them into their roles staying.

Leaders who ignore that fear won’t get the adoption they’re paying for. Asked what past technology waves taught them, one panelist named change management as the biggest factor in getting value from AI.

What I’d borrow from the CFO panel

AI adoption requires organizations to prove the value. The panelists admitted they’re still working out how to measure AI’s return. A few of their practices are worth copying:

• Set a baseline before go-live, then check whether results moved the way you expected.
• Separate hard dollars from capacity. Recovered revenue goes in one column, with hours returned, turnaround, and quality in another.
• Treat AI like a capital project with a shorter horizon. Five-year projections don’t hold at this pace.
• Tie today’s request to next year’s budget. One organization tells leaders that any dollar requested now becomes a commitment in their fiscal 2028 budget target.
• Price the opportunity cost of waiting, especially in competitive markets.
• Decide when you’ll stop. One panelist admitted healthcare is prone to sunk-cost bias. Write the exit criteria before the pilot starts.
• Run AI side by side with your people. In one shadow test, AI missed documentation a human coder caught in an unusual spot in the chart.

Education has to reach the whole workforce

Most AI training I see starts with clinicians. That makes sense for ambient documentation, but revenue cycle is where AI is already part of daily work, and where the fear runs highest. Coders, denial specialists, patient access staff, and billing teams need the same investment. In a 2026 HFMA survey, only about 7% of finance and revenue cycle leaders called their workforce “very prepared” for the revenue cycle of the future.

Here’s what role-based education looks like in revenue cycle:

• A shared foundation first. How AI produces its output, what data it can and can’t see, privacy rules, when to verify, and who to call when something looks wrong.
• Practice with the actual tools. Coders work AI-suggested codes in a training environment built from their own workqueues, and denial specialists edit AI-drafted appeal letters under supervised review. One speaker at UT REAL described pairing training with a week of hands-on practice, comparing performance with and without the tool, and checking review skills over time.
• Teach people to look for what’s missing. At CHAI’s fall symposium, a panelist described testing ambient AI tools by slipping a head injury into small talk, and some tools dropped it as noise. The revenue cycle version is the shadow test above. Reviewers need to check for omissions as well as errors to be kept on their toes.
• Protect how new staff learn. If AI handles the routine charts, new coders lose the practice that builds judgment. Keep some AI-free work in onboarding so people build the skill they’ll later supervise.

A session chair at UT REAL put the order as mindset, then skill set, then tool set.

Measure the education the way you’d measure the AI:

• A short confidence survey before go-live, then at 30 and 90 days
• How often staff accept AI drafts without edits, and how often they override them, checked against audit results, since a very high acceptance rate can signal over-trust
• Whether workload feels lighter, or just shifted from creating to reviewing
• Whether your revenue cycle team starts proposing its own AI use cases

That last one is a good sign the fear is fading. When people understand the tool and own the job of judging its output, they start talking about how their work changed.

Revenue cycle staff are the people best placed to judge what the AI gets wrong. Give them that job, and the training to do it well.

Our consultants are part of this workforce

The same shift is reaching Epic consultants. Cactus supports our consultants’ career growth, from placing them on projects that match their interests to backing their participation in conferences and educational opportunities.

If a reviewer for the Joint Commission’s new AI certification asked how your coders were trained on the AI they use every day, what would you show them?