Three-quarters of payer respondents say AI workflows need human review, explainability, monitoring, and audit trails before broader deployment. Trust is the gate between AI that’s deployed and AI that’s used, and it’s why Optura put trust on the agenda at In the Round 2026. An AI tool people won’t use has no value to measure.
Most organizations treat trust in AI as a single decision, made once the model is good enough. The Trust Equation panel described something different. Trust builds in installments, and each one makes the next, larger ask possible.
Start with an ask that needs no trust
The first installment costs nothing. A tool that leaves everyone’s workflow alone doesn’t require anyone to believe in it, so it can prove itself before it asks for anything. One organization began by pointing open-source tools at the thousands of public comments on proposed rules, and the people who saw it work became its first believers. That’s the job of the first step. It creates people who will vouch for what comes next.
The company behind one ambient documentation tool made the same choice with doctors. It built the tool into its product, gave it away, and didn’t require anyone to use it. Early users gained about six minutes per encounter, and same-day closure of those encounters rose about 30%. Doctors who got their evenings back told their colleagues, which no rollout plan could have done.
Borrow credibility before you spend your own
Trust in AI is social before it’s statistical. People believe people they already trust, which is why accuracy scores rarely move clinicians. When one health system’s models each wrote a patient handoff summary a little differently, the data science team’s numbers didn’t change minds. A nurse leader did, by showing that five expert nurses she’d trust with her life also wrote it five different ways. The evidence mattered less than who presented it. Find the person your skeptics already believe, and let them make the case.
Credibility also depends on being understood. Early in one health system’s work, an engineer and a change management leader spent minutes discussing “NLP” before realizing one meant natural language processing and the other meant neuro-linguistic programming. People can’t trust a case they can’t follow, so shared language comes before shared evidence.
Put your own credibility on the line
A leader who asks a team to use AI is making a withdrawal. A leader who uses it first is making a deposit. One company’s CFO spent a day building with AI what the team had been asked to deliver for six months, then shared it with a note. “If I can do this, you can do this.” Because the ask came from someone who had done the work, the team took it seriously, and it now handles reporting it had no capacity for a year earlier.
Let the believers recruit
Earned trust spreads faster through peers than through mandates. A federal agency’s early users now run an ambassador program that brings in the next wave, and the base of belief they built is what let the tool take on a far larger job, suggesting changes to final rule language. Each new believer lowers the cost of the next ask.
Then make the big ask, and keep the door open
The installments exist to make the large commitments possible. With clinicians and engineers aligned, the health system could fund a foundational AI platform with no straight-line ROI, and investment accuracy scores alone would never have justified. The CFO’s team is moving on to forecasting revenue for 170,000 clinicians.
Trust compounds in one more direction. At one theme park company, ideas that weren’t ready went into a greenhouse instead of a graveyard. Because people believed their ideas would be kept, an unwanted one was still there two years later, when it became the foundation for a digital version of a ride.
Trust can be spent down, too
The same compounding works in reverse. Asked which part of the AI chain they trust least, panelists split. Some pointed to the data, including output that comes back sounding more certain than it should. Some pointed to the money, with AI valuations running ahead of the markets they serve. And some pointed to the people reviewing AI, who tend to stop questioning it once it’s usually right.
That last one matters most for any organization counting on human review as its safeguard. Review protects trust only when the reviewer has the expertise, time, and authority to change the outcome. A review that has turned into a habit draws down the account without anyone noticing, until a mistake gets through and the trust built over months goes with it.
Where trust stalls
Most organizations ask for the last installment first, a large investment justified by a model’s performance, from people who haven’t seen it work in their own jobs. The answer is usually no, or a yes that never turns into adoption. Building trust in order is slower at the start and faster everywhere after. Optura helps health plans and health systems make the value of each installment visible, so every ask after the first comes with evidence behind it.
For the themes that ran across every panel and roundtable, read Full Measure: The Set List, Optura’s insights report from In the Round 2026.
