UNC Health’s road map for AI in revenue cycle: 4 insights

Two years ago, UNC Health shifted from experiencing isolated quick wins with AI in revenue cycle toward enterprise transformation, looking for ways that AI and automation could eliminate friction points for staff and patients.

The impact has been dramatic.

This year, use of AI and robotic process automation to support claim appeals has increased appeals throughput by more than 2,000 appeals to date, said Christina Slemp, DHA, MSHI, MHA, system vice president of revenue cycle-financial services for UNC Health.

“I think I’ve seen denials waiting 60 days to get looked at to going out the door within three days,” said Slemp, who joined UNC Health in July 2025.

AI for evaluation and management coding significantly improved coding accuracy for physician billing, which “gave us a huge financial lift,” Slemp said. Meanwhile, use of a patient-facing AI agent saved 2,000 contact-center hours in five months this year, handling 18,000 conversations without requiring escalation to a customer service representative. It also contributed to a 33% reduction in MyChart messages.

Slemp, who spoke at HFMA’s Annual Conference on “AI-enabled RCM: A practical playbook for digital transformation,” shared four lessons learned in using AI and automation to transform revenue cycle performance.

1. Build a governance-friendly culture.

“Governance should never be viewed as a barrier,” Slemp said. “When you build it up front, it allows you to move faster and quicker once it’s in place, and then, you’re not surprised; you’ve kept patient data safe; you’ve given patients better experiences.”

UNC Health formally launched a responsible AI governance framework across the organization in March 2023. Its governance committee includes representatives from across the organization, including the revenue cycle.

• Understand that AI can’t fix a broken process.

“When you lay AI over a broken process, it just exposes the process faster,” Slemp said. “You have to step back and understand the workflow: where the hand offs are, where unnecessary touches are being created, why people are doing things that no longer add value and what the organization’s policies and business rules are. If we automate that bad process, we’re just creating a bad outcome at greater speed.”

• Start with a business problem, not an AI tool that needs a use case.

“It’s very important to know what your friction points are and where problems exist before you look at AI technology,” said Slemp, whose organization has deployed 100 AI use cases across the organization. “Seeing the technology and then trying to make it fit into your system doesn’t always work.”

Ultimately, this process requires revenue cycle leaders to fundamentally rethink what the revenue cycle should look like, Slemp said.

“A lot of leaders aren’t thinking like this, and they’re missing the gold,” she said. “They approach AI from the lens of, ‘This is my revenue cycle; this is my system. How do I use technology to make it better?’ instead of reimagining what the revenue cycle should look like and what partners and tools they need to support that vision.”

• Don’t lose sight of the human element.

“When you’re exploring where to add technology in the revenue cycle, don’t just think about the financial ROI. Think about how it could make the patient experience better and whether it will make a difference for your team,” she said.

“If the technology doesn’t fit the employee’s natural workflow, if the end user doesn’t trust it or if it creates another touch somewhere else, you’re not really solving anything,” she said. “Simplify first, automate second and then measure the impact for patients and staff.”