Full Measure: What It Takes to Agree on AI’s Value

“Let’s add AI as a fifth strategic pillar.”

“This will free our people up for higher-value work.”

“We only have 37 AI projects in flight.”

“Don’t worry, it’ll be on our vendor’s roadmap.”

For anyone running a healthcare organization, these lines sound familiar. They sound right, and they’re also what people say when there’s nothing real to measure. 

A measure means two things. In business, it’s how you prove something worked. In music, it’s an agreement on the shared count that a band plays together. Healthcare AI is short on both.

Optura’s In the Round 2026 brought together the people who define AI’s value in healthcare, from every seat, to share experiences and perspectives that improve care and the ecosystem around it. On a farm outside Nashville, the event blended music, nature and industry under one theme, Full Measure, named for the shared count in music that lets everyone play together.

Some leaders plan on three- and five-year horizons. Others answer to quarterly EBITDA and patient satisfaction scores. Inside one organization, the CFO watches the P&L, the technology team watches risk, the program team counts what’s moving, and the people closest to patients and members want to know whether anyone’s day got better. Each of them is right, and each is measuring something different.

The organizations getting real value from AI have settled what “this is working” means. Project count has little to do with it. They’ve had the hard conversations about which value drivers come first. Total cost of care or administrative cost reduction? Clinician satisfaction or patient satisfaction? Once those answers exist, everyone knows what to measure. Leaders have grounds to shut down what isn’t working, and the CFO can say what moved, by how much, and who owns it.

AI raises the stakes, because AI pursues whatever score it’s given. It beat the best humans at chess, then at the board game Go. Optura co-founder and CEO Andy Fanning pointed to an AI playing Super Mario Bros. that learned the game awards points for forward progress and started stepping back and forth to rack them up. That’s reward hacking. When a measure becomes the target, a system will find ways to hit the number without doing the work the number was meant to capture. Without an agreed score, AI optimizes for whatever it finds. 

None of this is exotic. Healthcare organizations already measure this way in other parts of the business. What’s been hard is doing it together, when every function has its own definition of what counts.

In a band, everyone counts the same beats and starts at the same time. The count doesn’t tell anyone what to play, and the fiddle never covers the guitar part. Every function in a healthcare organization should keep playing its own part. What’s been missing is the shared count, the full measure.

Most boards don’t have deep technical knowledge, and groupthink fills the gap. Leaders described educating directors along the way, and starting early, until the board understands how AI works well enough to weigh in on strategy. Picking use cases stays with management.

The education can’t stop at the board. AI brings a business change with it, and the mindset shift that comes with changing how work gets done reaches every level. Roles are shifting and new jobs are appearing. Workflows must change to make room.

Healthcare starts from behind. Plenty of core processes still run on fax machines, and margins are thin enough that efficiency isn’t optional. The same gap is why the potential is so large, and why technology partnerships matter. They let an organization leapfrog instead of catching up one step at a time.

Several leaders also described a committee problem. Committees tend to form around problems people anticipate. The useful ones are built to solve a problem that already exists.

One panelist quoted a patient safety advocate: “Clinicians move at the speed of trust. Patients move at the speed of desperation.” Panelists saw the same pattern with staff, because people who are drowning will take any tool that makes the day better.

That’s where trust starts. Providers using ambient documentation are happier, and their coding is more accurate. In small practices, AI can flag what needs validating before a well visit, work a physician used to do at night with a paper chart and sticky notes. That kind of adoption only happens when tools are in people’s hands and leaders believe in the strategy behind them.

Trust also comes from showing. Clinical colleagues were uneasy that different models wrote the same patient handoff differently. What won them over was a nurse leader asking five expert nurses to write it. Every version was different, and every one was trusted.

There’s no AI strategy without a data strategy. AI is only as good as the data available, and much of how an organization runs isn’t written down. The context lives in the heads of people who have done the work for 30 years. When AI can’t connect the dots, it makes assumptions, and bad or incomplete data multiplies the error.

Agentic AI raises the bar. Agents need context to make small decisions along the way, and much of that context sits in operational detail most organizations don’t treat as data. People filter out what doesn’t matter without thinking. AI can’t comprehend like that yet.

Asked which link they trust least, panelists split. Two named the data, including AI output that comes back overconfident. One named the money, with valuations running ahead of the market. One named people, because reviewers in the loop tend to stop questioning. The investor’s bet was on founders who understand both healthcare and technology.

Panelists debated whether every organization needs a chief AI officer. One pointed out that nobody has a chief electricity officer.

Ownership works best in pairs, with an executive who owns the return and a doer who owns execution. The doers can come from anywhere. One of the strongest examples came from a lawyer who built her own system to triage incoming contracts and check redlines against standard responses. Curiosity and urgency matter more than title.

Whoever owns the number has to understand the model well enough to stand behind its decisions. One prior authorization automation saved clinician time on paper and made a strong case study for leadership. The clinicians never fully bought in, and nobody owned turning the saved time into value. Fewer staff? More authorizations handled? Nobody could say.

Return on investment starts with the outcome you want, then the measures that show you’re getting closer. They also force discipline on spend. One leader had teams walk through their token usage, use case by use case, at 6 p.m. every day for a month so it would become a habit. Over the next two years, that attention cut the cost of their first use case, drafting replies to patient messages, by about 99%. Learning counts as a return too, especially when it shows up later in retention or client loyalty.

Agents need managing. Until now, software behaved the same way every time. The first time one team’s orchestrating agent fired a subagent, they realized they needed expectations and checklists, the same structures that keep people on track.

The workforce is changing with it. New graduates arrive already fluent in prompting and directing AI. One panelist described approving no new headcount for any department until its leader presented an AI plan. Another is starting to hire for the ability to direct a fleet of agents. Performance reviews can ask how people changed their jobs with AI.

Consumers are moving faster than the organizations serving them. In a 2026 ZS survey, 89% of U.S. consumers who use AI for health information found it helpful, close to the 96% who said the same of their primary care physician, and 68% of physicians reported more patients asking for a specific therapy by name (source). One payer on stage is already watching members connect their own AI agents to its member portal and deciding how far to let them in. Patients are arriving with a working diagnosis while most organizations are still deciding how far to let AI into clinical work.

It starts with believing there’s real value in AI and taking the time to think about where. Fear slows everything down. Panelists discussed how they are now bringing their tech teams into the room with the problem itself and asking how to solve it, so tech isn’t left taking orders on what to build.

The organizations moving fastest build a culture of ideas and keep learning. The path is never linear. Sometimes the right move is to blow up the current workflow and redesign how the work gets done, especially with physician and nurse shortages that more headcount won’t solve.

Failing fast is hard to accept in healthcare, where failure can mean life or death. That instinct has a cost. A lot of AI work starts, lingers, and never graduates to production.

Small, meaningful wins help. So does building alongside the people affected. Staffing and scheduling are emotional topics for frontline teams, and the tools that work are the ones co-created with them before they scale.

AI is moving too fast for any leader to have the answer, and the leader as subject matter expert doesn’t hold anymore. Several leaders described letting go of that, and admitted it’s still hard. The job now is to set the outcome and create the conditions for teams to find the how. Teams handed a problem to solve often come back with a better answer than the leader imagined.

Judgment matters more as a result. Leaders still have to challenge inertia and push their teams to do the hard thinking. They also need time to understand what the data is telling them. Technology was supposed to free up time. Most leaders are now on 24/7.

Slowing down is the hardest part, and saying no is the hardest part of slowing down. Arthur C. Clarke wrote that “any sufficiently advanced technology is indistinguishable from magic,” and teams are producing small magic moments constantly. Most of them aren’t on the priority list. The discipline is to say that’s great, then return to the work tied to outcomes.

AI doesn’t have a job description, and it won’t be held accountable for the outcome. Accountability starts and stops with the leader, whether a decision came from AI, a direct report, or someone three levels down. With this much data and speed, leaders still have to invest in people, develop them, and surround themselves with others who are curious.

• Before you start, decide where the impact should show up and how soon you’d expect to see it.
• Work on what only you can do.
• Check what you’re actually rewarding. Goals built on milestones and effort pull teams away from outcomes.
• Take one day a month away from the office to think.
• Treat anything that doesn’t affect the patient or member as waste.
• Thank the skeptics. They will find the hard things and then where things break.
• Celebrate small wins that matter, even before they’re big ones.
• Be honest about what you don’t know. People follow a leader who admits it.
• Nobody can see the future. Map the range of possibilities and build an organization resilient enough to absorb the bumps.

You can shape AI or be shaped by it.

Shaping it starts with agreeing on what you’re trying to measure, and counting it the same way across the organization.