The question consuming healthcare boardrooms is no longer whether to invest in AI. That debate is over, and the capital is already moving. The harder question is one of returns: even as 81% of health system C-suite leaders plan to grow their AI investment in fiscal year 2026, according to Huron's 2026 Executive Provider research,1 few can articulate what their previous investments actually delivered.
The stakes of leaving that question unanswered keep rising. Nearly half of executives are actively preparing for federal reimbursement cuts, margins remain compressed, and every technology dollar competes with clinical and workforce priorities. Yet AI ranked as the industry's top strategic concern, top leadership challenge, and top investment priority in the research, which surveyed 300 executives across national, regional, and academic health systems. The real issue is why the value so often fails to materialize where the leaders funding it can see it.
The research offers a clear diagnosis: senior leaders and the VPs and directors they work with disagree about whether their AI initiatives are paying off. Nearly a quarter of C-suite leaders say none of their AI initiatives have generated measurable ROI. Ask the VPs and directors running those same initiatives and the picture flips: not one reports zero return. From where they sit, the wins are visible: hours recovered in a clinic, claims moving with fewer touches. But most of those wins never get a baseline, so they never roll up to the ledger the C-suite reads.
When AI frees up clinician hours or takes touches out of a claim, the time savings does not convert itself. Someone has to convert that gain into something measurable, such as more visits, lower staffing spend, or reduced clinician turnover. In our work with health systems, we often see this opportunity stall because responsibility is never formally assigned to the right owners: the leaders who have the authority to redesign workflows and reallocate resources, and the willingness to make the hard calls that conversion demands, such as consolidating shifts or reshaping roles. Too often that mandate falls through the cracks because ownership is unclear and the authority to act is fragmented across multiple stakeholders. Freed capacity quietly reverts to slightly lighter days rather than measurable margin impact.
An unassigned responsibility to the right owners is a solvable problem, which makes this diagnosis more encouraging than it sounds. Some AI solutions underdeliver, and some deployments should rightfully be shut down. But technology failure is rarely the whole story. More often, the missing ROI traces to a value target that was never given an owner, a baseline, or a destination. ROI is simply value that has been measured all the way to the ledger. Closing that distance is a matter of leadership, not technology. The health systems that pull ahead will be the ones that treat it that way, with intentional decisions around which AI solutions to scale and which to decommission and holding the right individuals accountable for the value delivery.
Capacity is not value until someone converts it
In practice, converted value takes one of three forms in a health system, and each has a natural owner on the executive team. Recovered capacity can become revenue growth: added visits, added procedures, and shorter backlogs. It can become lower premium-labor spend, with agency, overtime, and locums retired. Or it can lead to reduced clinician and employee turnover.
Naming the owner before go-live gives the recovered hours a purpose from day one, with someone accountable for turning them into a specific result. They become added visits, retired agency shifts, a clinician who stays, or fewer after-hours charting sessions. When no one is named, the hours get absorbed into business as usual and never show up on a financial statement.
One COO interviewed in the research described how that outcome looks from the budget: "Here's the dilemma. How do we pay for it? We're doing the same amount of work, and we just added a bunch of expenses." That is not a poorly run deployment. It is a successful one with no conversion designed. The technology delivered, and the organization still cannot find the return.
That view is common at the top. When leaders grade their full AI portfolios, 67% of C-suite leaders say fewer than a quarter of their initiatives are generating returns, while only 21% of VPs and directors see it that way. The people closest to the work also have an explanation. Asked what is holding results back, VPs and directors point less to the technology than to whether clinicians and staff actually adopt it.
The pattern has a precedent, and it is the one we share with boards. Edison lit lower Manhattan in 1882, and American factories electrified quickly. Yet measured productivity barely moved for roughly three decades, because factories swapped the steam engine for one large central electric motor and kept everything else: the same belts, line shafts, and multi-story layouts built around a single power source. The gains arrived in the 1920s with unit drive, a small motor at every machine, which let factories reorganize around the process rather than the power source. Most health systems are in that central motor phase today. Ambient documentation gets dropped into an unchanged clinic template. A coding engine gets layered onto charge capture that runs four different ways across four hospitals. The AI solution works. The factory has not changed yet.
Run AI like a portfolio, not a collection of pilots
The solution is less exotic than the problem. It looks mundane on paper, which is exactly why so few systems do it. It starts with one list detailing every AI project in the organization, including department-level pilots, vendor features quietly switched on inside the EHR, and undocumented solutions staff already use on their own. Most health systems cannot produce that list within a week. That inability is usually the first thing we discover when we begin working with a system.
Once the list exists, the real discipline begins. Every project on it carries three requirements: an executive who answers for the result in their own budget, a success measure recorded before the solution goes live, and a date when leaders will decide to expand it, rework it, or end it.
Some of AI's returns show up as dollars, and some do not. Both deserve the same rigor. The CFO owns the dollars: revenue cycle automation, prior authorization, patient throughput, and reduced agency and overtime spending. The CMO, CNO, and CHRO own the outcomes that resist a price tag: clinician wellbeing, patient safety and experience, and competitive standing. Retention shows how the two ledgers meet. The CMO and CHRO own whether clinicians stay, and every clinician who does appears on the CFO's ledger as a replacement cost avoided. Hard to monetize is not the same as impossible to measure. Each outcome still gets a starting measurement, a way to track it, and a regular review. Boards already treat the two as equals: in our research, competitive positioning, financial ROI, and patient safety carried nearly identical weight on board agendas.