Aetna CTO talks prior auth AI strategy, AI governance

Keeping AI out of clinical decisions, denials

AI's role in clinical decision-making -- or, as Frank emphasized, lack thereof -- is a foundational principle in the payer's overall AI strategy.

"AI is not going to deny care," he stressed. "It's not going to make clinical decisions on the behalf of our colleagues, and we've stayed true to that over time."

Instead, Aetna adheres to a human-in-the-loop philosophy.

Take prior authorizations, Frank offered as an example. Aetna uses AI in its prior authorization process, meaning the technology adjudicates electronic prior authorizations and issues approvals. According to Frank, the payer's AI approves around 82% of prior authorization requests in real time.

When the approval isn't straightforward, meaning Aetna doesn't deem AI the appropriate adjudicator, the request goes to a human reviewer. AI has a role to play here, too, Frank said, but that role isn't hitting the denial button. Instead, AI is responsible for drawing in the necessary information for the human to make a decision.

That's the same playbook a lot of payers say they're using, but some industry professionals say the process is more complex than that. Physicians, for example, are alarmed by the use of AI in the prior authorization process, according to 2025 surveying from the American Medical Association.

Experts say AI prior authorization review is not a black and white issue.

"I want to go on record that I'm not suggesting physicians are biased with this. What I'm saying is that technology determines which cases enter the denial risk pathway," Stacie Sutter, vice president of payer strategy at Ensemble Health Partners, told our sister site RevCycle Management. "The computer may not write the denial itself or the denial letter, but it often helps determine which of these cases flows through as not high risk versus which ones pause for provider review."

Still, Frank said Aetna has never wavered on its human-in-the-loop approach to AI utilization, and he doesn't think the policy will change anytime soon.

"It's just core to what we do every day," he explained. "I think you're going to see as agentic AI starts to mature, people will push the boundaries of what AI can do in the workflow, and I think it's really important to have those guiding principles upfront."

Using AI to streamline payer staff workflows

While Aetna claims a strict "no AI in prior authorization or care denials" code of conduct, the payer is still realistic about how AI can benefit the business.

Like many in healthcare, Aetna views AI as a critical tool for streamlining clinical workflows, especially for healthcare professionals bombarded with growing workloads. However, Aetna takes a product-first approach to AI, seeking to eliminate a business problem rather than deploy AI for AI's sake.

"Our goal is around matching up AI with humans, and we think we get the best results when we do that," Frank said. "We're trying to use AI to remove the friction, whether it's for our members, our providers or our colleagues and ultimately that drives where we think AI can really make the most difference in the industry."

Aetna has its clinicians and care managers use ambient AI, just like a doctor might use the tool on the hospital or health system side. The technology lets Aetna's care managers consolidate key information before meeting with members and view AI-generated notes at the end of a virtual visit.

"That's a win for the colleague who can spend more time with the member, but it's also a win for the member who actually gets a one-on-one personal experience versus being rushed or having to wait while care managers document," Frank said.

Aetna has also implemented AI that helps members find a clinician who's accepting new patients by leveraging claims data and flagging providers who had a new patient within the past three months. The payer is also working with AI tools that promote price transparency, letting members get a cost estimate within the context of their specific plan, provider and service.

Experimenting under AI governance

At the core of Aetna's AI strategy is the idea that the technology should not be deployed simply for show. Frank emphasized that AI needs to solve a tangible problem for the company in order to see the light of day.

It's that mindset that's helped Aetna introduce these products to a workforce and member population who might be reticent to trust the technology.

"We've created a unique culture at Aetna and CVS Health where we're a product-oriented company," Frank noted. "When we think about these AI agents, we're not just thinking about deploying an agent and having a bunch of agents run around. We think about what the product is that we're going to create that's going to solve that friction point."

That's led to a robust AI development and governance strategy that centers around the colleague, Frank added. For every AI agent that's introduced, it's been shaped by relevant employees who will actually be impacted by the AI. For example, Aetna looped in call center leaders when it designed its product in what the payer calls pods.

Pods include members of the impacted team, plus product managers from technology, engineering leads and designers. By having subject matter experts from the part of the business impacted by the AI, Frank said teams are more receptive to the tool, Then, the AI has a greater impact.

Aetna also has clear mechanisms for team member feedback to ensure the AI is working correctly and yielding accurate results.

This all fits inside Aetna's greater AI governance strategy, Frank added. The payer's AI governance team includes the chief information security officer, compliance and regulatory team members, engineers and product managers. The organization also taps business subject matter experts who review every case.

Still, the payer has built a backlog of AI use cases, and Frank doesn't want it to take 90 or more days for teams to get their AI requests approved.

"We've actually built AI into the governance process," he said.

The AI for AI governance is designed for use cases that fit a pattern at Aetna and are low-risk. If the tool is going to be similar to another initiative, it might get a faster automated approval.

"If you're going to use the same guardrails, the same techniques, that should be a really fast approval, and then obviously we'll measure and monitor the outcomes," Frank said.

Sara Heath is an executive editor at Xtelligent Healthcare Media, where she covers patient engagement, healthcare policy and health IT.