What to tell your board about agentic AI

For the past two years, the pressure has come from the top. Health system CEOs told their CFOs to “adopt AI.”

The problem: Generative AI, made famous by ChatGPT’s launch in 2022, is based on a pure large language model approach that runs into real limitations in healthcare finance. It doesn’t know how your organization defines contribution margin, or how your clinics roll up to service lines.

CFOs were being pushed toward a tool that wasn't built for their domain. It was pressure without a credible path forward.

That dynamic is shifting…and faster than most people expected.

What changed

Agentic AI—where the technology has gone since ChatGPT—is a different category of capability than generative AI. A large language model answers questions. But an AI agent does autonomous, end-to-end work spanning systems and making decisions at a cadence that doesn't depend on a human initiating each step.

More importantly: agentic AI that's built on a context layer, one that encodes both industry-wide rules and organization-specific logic, produces outputs that are trustworthy enough to act on. Trustworthy enough, as it turns out, to put in front of a board.

The conversation has shifted from "can we use AI in healthcare finance" to "here's what it's actually doing in our operation." That's a different board conversation. And CFOs who recognize the moment have a window to lead it, rather than react to pressure from above.

What it looks like at scale

John Orsini, EVP and CFO of Northwestern Medicine, described the board dynamic at a large academic health system the way you'd expect from someone who's been careful and deliberate about every major technology decision:

"[It] was much easier to have a discussion around agentic AI. Some of our board members who are in private equity or large companies have encouraged us to really take this on in earnest because they've seen benefits in their organizations."

The education burden, in other words, was lower than expected. The more extensive work was internal: aligning with the CIO, getting clear on what existing vendors were already building, and identifying where an independent capability—one that functions like a "super financial analyst," in Orsini's words—could do something none of those systems could.

"It can't be a nice-to-have. It can't be, ‘well, this makes my life easier because it's providing insights’....  But if nobody's doing anything with it, we'll shut it down. We probably need to find $1.20. And we hope that it'll be two, three, four, five, ten dollars."

What it looks like at a critical access hospital

John Everett, CFO of Wray Hospital and Clinic, a critical access hospital in rural eastern Colorado, tells a different story. Not a deliberate, phased evaluation, but something closer to a snap decision.

"It took a matter of: here's the quote, they talked for five minutes, motion to approve, done."

What made it that fast?

"The bigger thing out of it wasn't really having to talk about AI. It was having them understand what the system would do. And so they're like, ‘Oh, we've been asking for that for years.’"

The board didn't approve an AI strategy. They approved a solution to a problem they'd been watching accumulate for years: data that was six months old before it was usable, a finance team of three people managing month-end close entirely by hand, department heads making decisions without visibility into their own P&Ls.

When the problem is that visible—and the fix is that concrete—the “AI conversation” almost doesn't need to happen.

What to actually do in the next 90 days

To get ready for AI, the most useful question isn't which vendor to call. It's what to do before that call. Everett was direct:

"Know where all your data lives and how to get it out of all of your systems. Know what you actually want an agent to do. Have that strategic plan written down before you start."

Orsini added two things: get aligned with your CIO before you go looking at tools, and take stock of what your existing vendors—your ERP, your close platform, your analytics stack—are already rolling out. Understand the baseline before you evaluate what's adjacent to it.

The CFOs who move on this now aren't taking a risk. They're filling a vacuum…and walking into their next board meeting with an answer to a question that's been sitting on the table for two years.

‍