From Mandate to Muscle Memory: Notes from the room.

By Madeline DuPlessis, SVP, Enterprise Sales

Raise your hand if AI is part of your day-to-day work. Keep it up if you can name a measurable outcome from that AI in the last quarter. Keep it up if a team outside your own could name that same outcome and agree on what drove it.

I put those questions to the room on Sept. 29 at Health Evolution Connect in Nashville, where I moderated “From Mandate to Muscle Memory: Earning Trust Beyond the C-Suite,” a Brass Tacks panel. It was a small room, so we kept it interactive. Most hands went up for the first question. Fewer stayed up for the second.

The hardest part of enterprise AI comes after the contract, when leadership hands a tool to the organization and expects adoption to follow. Adoption follows trust, and trust has to be earned at every level below the person who bought the tool.

I told the room that patients are warming up to AI faster than the organizations that serve them. In a blinded study Google published in Nature, patient actors rated its diagnostic AI, Articulate Medical Intelligence Explorer, above primary care physicians on 25 of 26 measures, empathy included.

The fear has shifted to the people inside healthcare organizations. Is this going to take my job? Am I making the right bet? What if the agent goes rogue? Are payers about to deny everything because a bot is doing the work? Those questions sit with operators and leaders.

After years at IBM, Oracle, and Google, I know how much trust a logo can carry into a room. A startup earns it one meeting at a time, which makes it easy to see where trust breaks. Most often, it breaks after the purchase, somewhere between the executive sponsor and the people expected to use the tool.

Tom Cassels, managing director at Manatt Health, joined me on the panel. He spent the past year studying health systems that moved AI out of pilots and into production. In those systems, he said, CEO-led strategy worked as long as the CEO stayed involved. Some of the biggest clinical AI impact came where the CEO was a clinician, and trust held when staff saw their leaders using the tools themselves.

“When a strong opinion, strategy, or communication was made and then leadership sat back and watched the implementation teams struggle, that’s when there was just an immediate break in trust,” Cassels said.

After a break like that, even the next initiative is harder to lead.

Kristen Valdes, founder and CEO of b.well Connected Health, who was also on the panel, has been in rooms where the C-suite signs a contract and delegates. Those projects “always, 100% of the time” go slower, she said, and are usually off track and off budget. There are always people pulling in the opposite direction to make them go away. A common variant is the consensus spiral. Five leaders agree the project should move, then someone decides 10 more need a seat.

“It’s just this death by everyone has to agree, and that’s just fear,” Valdes said.

Ownership reverses it. Valdes described one project that moved from a delegated model to a C-suite-driven one.

“Something that we were unable to accomplish in two years, we were able to get live to market in under four months,” she said.

In her experience, that works when the C-suite sets a hypothesis and a deadline and tells the team it’s fine to be wrong as long as they learn quickly.

I gave the room an example of a team that spends its days answering RFPs. Leadership sees an obvious use case: an agent that reads each RFP and drafts the response. The RFP says it’s due in seven days. The agent turns it around in six. Leadership counts a day saved.

The people who have done this work for 20 years know what the cover page doesn’t say. That buyer usually picks whoever responds first, often by day three. Six days is still late. The agent is useful, and leadership is grading it against the wrong number.

When leadership and the people doing the work disagree about what counts as value, a tool can hit every target in the business case and still look pointless to the team using it.

The advice I gave the room at the close was to agree on value drivers before anything launches, and to agree across every level. That means the executive sponsor, the operators, the end users, and, in healthcare, often the patient or member. 

That agreement is where ROAI™ (Return on AI Investment) starts at Optura. Leaders set the value drivers first, and performance gets tracked against them continuously, with no waiting for the postmortem. When every level helped set the drivers, the return at the end is a number the whole organization believes.

Ask a team whether they believe in a change leadership told them to make, and the answer is yes every time.

“If you told me to try it, I’m not going to tell you that I don’t trust it,” Cassels said.

Are people running the old workflow with an agent in their way? Or are the people closest to the work already ahead of the AI, changing how the job gets done? A stall can mean the organization solved the wrong problem, and that only becomes visible to someone who watches long enough.

Patients follow the same pattern. At one client’s clinic, Cassels saw signs on exam room doors asking patients to leave their ChatGPT outside. I asked the room to guess how many of those patients didn’t come back for their next appointment. The guesses were 80 and 90 percent. It was 60 percent. Cassels’ read: they went looking for a doctor who would engage with the research they’d done. Employees do the same with a tool they don’t trust. They route around it.

Valdes gave the room an example from scheduling. A front-desk scheduler knows one surgeon operates on right hands Mondays and Wednesdays and left hands Tuesdays and Thursdays. Rules like that have broken scheduling decision trees for years. Today, an agent can be built from how she does the job, which makes her the person who trains it.

“What you really need is Betty at the front desk,” Valdes said. “You don’t need better, smarter engineers and product managers that are writing the requirements.”

That’s why the most important trust in an AI program often sits in the middle of the org chart. AI councils and champions work below the executive sponsor and above the end user, and their colleagues already believe them. No vendor or model card has that kind of credibility.

One more decision separates the organizations pulling ahead: naming what good enough looks like. Waiting for a perfect model is its own way of stalling. Once a model ships, it still needs watching, because AI in production drifts from what was tested. Models and data both change, and the monitoring to catch that drift is part of the cost.

Cassels framed leading from the top as clearing obstacles for the people doing the work, human and agent alike. Do that consistently, and the label fades.

“We are just going to be companies and we’re not going to talk about AI the same way we don’t talk about the internet in how we operate our businesses,” he said.

That’s muscle memory.

If your organization is working out what its AI investments are worth, talk with the Optura team.