Health systems are spending heavily on AI, but showing the return on the investment remains a challenge.
During a featured session at Becker’s 14th Annual CEO + CFO Roundtable on Nov. 3, two executives from Premier — a technology-driven healthcare improvement company — argued the problem is less about the technology and more about the discipline around it.
Florian Quarré, Premier’s chief AI and analytics officer, oversees Premier’s AI and related technology investments. He works with Premier’s CEO, CFO and board to measure the value of the investments. Matthew Zubiller, Premier’s chief business and product officer, has spent nearly 30 years in healthcare on both the payer and provider sides, including revenue cycle and quality.
Note: Quotes have been edited for length and clarity.
1. Don’t limit AI’s value to efficiency
Mr. Quarré said one of the most common mistakes he sees is treating AI only as a tool for cutting costs. “Limiting use cases to just FTE reduction and just efficiency creation alone is oftentimes one of the biggest traps that we’ve seen happen,” he said, referring to full-time-equivalent positions. AI can also guide better decisions, improve quality and manage risk, he said, and organizations should count that value too.
Mr. Quarré also cautioned against equating AI with generative AI. “There’s still a lot of what we now call more traditional machine learning and all of those expert systems that are extremely relevant,” he said.
2. Map it, measure it, then move
Mr. Quarré suggested a three-step framework. First, map where AI is relevant across the organization. Then measure against a baseline. Finally, carry out a prioritized sequence of moves. “If you don’t measure, you don’t know what you’re going to do,” he said.
He used document digitization at Premier as an example. The same use case touches finance, supply chain and clinical records, Mr. Quarré said, and Premier measured how much time staff spent on the manual work. Organizations can work in small pieces, such as adopting ambient AI documentation or coding support, and repeat the cycle, he added.
3. AI needs change management to deliver a return
Mr. Zubiller said Premier believes the return on investment from AI has generally fallen short, for many reasons. “The reality is that AI on its own is not good enough,” he said. “You need to be able to have change management, you need to have appropriate controls in place.”
He contrasted digging through multiple reports with asking a plain-language question, such as where a system can find opportunities to reduce supply costs for cardiac rhythm management devices. In Premier’s example, the tool returned opportunities ranked by dollar value. “You have to be able to have an interface where you’re asking this question as a user, not having to have the deep expertise to be able to know how to ask the question,” Mr. Zubiller said.
4. Clean, unified data comes first
Mr. Zubiller described a health system CFO who got different answers to the same question depending on whether he asked supply chain, procurement or the medical side. To produce reliable results the data has to be contained in one way, in one single infrastructure, Mr. Zubiller said.
Mr. Quarré made the same point from the model side. “You can’t be running AI without the data that is going to be relevant,” he said.
5. Treat AI as a portfolio and hold vendors accountable
Mr. Quarré said a portfolio approach to AI investment is critical since some AI bets fail. He encouraged executives to press vendors on whether their data will be used to train vendors’ models, how risk is shared if records leak and what service levels and turnaround times apply.
Mr. Quarré urged leaders to move from reactive to intentional AI use. “Define the value before you go at it and measure it twice, cut once,” he said.