5 lessons on healthcare AI ROI from leaders at ViVE 2026

The “Crossing the AI Chasm” panel at ViVE 2026 opened with two questions for the audience—”What is your goal for AI in clinical use?” and “If AI achieves that goal, how does it affect the finances of your organization?” Farzad Mostashari, Aledade’s CEO and co-founder, argued that for most people in the room, the answers do not align. The former national coordinator for health IT has been making this case for more than a decade.

Mostashari pointed to the EHR era as a cautionary precedent. Adoption alone, he argued, does not drive clinical improvement. EHRs spread rapidly across U.S. health systems over the past two decades, reaching nearly every hospital and most physician practices. The EHR features that ended up most developed were the ones that generated revenue under fee-for-service reimbursement, including documentation, coding, and billing. Population health management tools did not get the same investment. 

The same risk applies to AI tools. The ones most likely to scale will reinforce existing revenue, not change clinical outcomes. Five lessons from health system leaders at ViVE 2026 show how to break this pattern.

ROI depends on which leader is measuring

How ROI is defined depends on who is asking, says Abdul Shaikh, AWS’s global leader for population health. CFOs measure cost savings, chief medical officers measure clinician burden, and population health leaders measure value-based and quality metrics. AI tools that only satisfy one leader’s metrics deliver narrow wins that do not scale.

Michael Han, CMIO at MultiCare Health System, ran an ambient documentation deployment that satisfied all three. Hard ROI came from productivity gains and improvements in HCC and RAF coding accuracy. Soft ROI came from a clinician burnout drop from 60 percent to 16 percent over the trial period, a result Han said he had not seen any single intervention produce in 15 years of informatics work. Burnout reduction translated into retention dollars, which moved it onto the CFO’s ledger.

Projected ROI only counts if the budget reflects it

Stuart James, deputy CIO at Christus Health, shared a Dragon transcription deployment from earlier in his career. The chief nursing officer projected $2 million in transcription savings to win approval for the software. Two weeks later, in the budget meeting, she objected when the IT team tried to cut her transcription budget by $2 million. She wanted the technology and the transcription budget. James’s point was that ROI claims used to justify AI purchases rarely match what organizations actually book, because no one ever cuts the line item the savings were supposed to come from. The AI runs, the invoice arrives, and the original cost stays in the budget. The CFO sees a new expense without an offsetting cut.

Abundance use cases generate ROI that scarcity workflows cannot

Healthcare has always been designed around clinical scarcity, with triage and risk stratification used to narrow outreach to the patients most at risk because health systems do not have the staff to contact everyone. AI can remove this constraint, explained Munjal Shah, CEO of Hippocratic AI. Last summer in New York, his company’s voice agents called 16,000 Medicare Advantage members during the hottest four hours of each day of a heat wave, conducted a heat stroke assessment in the patient’s preferred language, and dispatched a ride to a cooling center—a public, air-conditioned space—when needed. The entire program cost less than a single ER admission for heat stroke would have. Shah estimated this same outreach in human effort, and determined it would have required 4,000 clinicians, with labor costs up to $2.8 million. 

In a separate deployment, his company reached 250 of 1,700 patients lost to follow-up on lung nodule scans. The health system had already tried text messages, MyChart notifications, and physical letters before turning to the AI agents. The calls led to one patient’s life being saved while generating $2 million in CT, biopsy, and cancer treatment revenue at $2,000 in agent costs. In both of Shah’s examples, the revenue followed the clinical outcome, not the other way around.

Vendor pricing should align with the outcome

Jennifer McCraw, vice president of enterprise access and digital transformation at MedStar Health, described the contracting structure behind the system’s care navigation deployment with Be.Well. The vendor charges per interaction rather than a flat license fee, and the contract is structured so both organizations are rewarded for growth in patient bookings.

MedStar reported a 77 percent increase in provider search click-through rate and a 50 percent increase in bookings after switching to the LLM-powered tool. The pricing structure made the deployment economics work for both parties. When bookings rose, both organizations benefited, and when usage fell, neither did. Flat-fee licensing produces the opposite incentive, paying the vendor the same whether the tool is used heavily or sits idle.

The cleanest ROI is when the business model already rewards the outcome

Mostashari illustrated his thesis with an example from Dr. Jeremy Presley, an independent primary care physician in Dodge City, Kansas. Presley’s practice piloted an AI agent from Mostashari’s company that calls patients overdue for wellness visits. The agent talks with patients in their preferred language, calls at whatever hour is convenient for them, and is paid for out of Aledade’s share of the savings the practice generates. Since the AI tool produces savings by keeping patients healthy and out of the hospital, it pays for itself.