As AI transforms healthcare operations, the industry’s most consequential challenge has shifted from building models to deploying them responsibly in regulated workflows where trust, accuracy and human judgment must coexist. This is where human-centered AI healthcare becomes essential, ensuring automation supports faster decisions without displacing the expertise and empathy required in care-related workflows.

The challenge is especially acute in payer operations, where complex administrative functions can directly affect how quickly providers respond to members’ needs. Notably, provider inquiries, claim research and correspondence management remain labor-intensive, multi-handoff processes that delay care, according to Danny Brakebill (pictured, left), staff vice president of business operations at Elevance Health Inc., a health insurance and benefits company.

“We took … one of our processes that has a lot of handoffs, requires a lot of research and also is kind of painful,” Brakebill said. “Instead of [providers] having to recreate a claim and understand what happened to answer a question, we get them 90% of the way there and they can just answer the question quickly and get back to providing care to our members.”

Brakebill, Venkat Alladi (center), staff vice president for technology at Elevance Health, and Balaji Ramdoss (right), managing director at Deloitte Consulting LLP, spoke with theCUBE’s John Furrier at Google Cloud Next, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how Elevance Health is deploying human-centered AI healthcare solutions to simplify provider operations, accelerate time to value and build responsible AI governance from the ground up. (* Disclosure below.)

Human-centered AI healthcare built on trust and responsible design

To address the challenge of labor-intensive, multi-handoff provider inquiries, Elevance Health and Deloitte partnered to build a solution on Google’s Agent Development Kit, an open-source framework for deploying multi-agent workflows at enterprise scale. The approach synthesizes unstructured incoming provider data and surfaces pre-assembled context for associates — reducing manual effort while keeping human judgment at the center, Alladi explained.

“When we looked at this problem, we couldn’t look at it [in] a traditional way,” Alladi said. “The problem brought in a lot of unstructured information that we had to synthesize and we needed new tooling or technology, such as AI, to bring all of that information together, look at different formats of information coming in, understand what it meant and then come up with a scalable solution with security and governance in play.”

Responsible AI governance was not an afterthought — it was architecture. Legal, compliance and security teams were embedded in every design review, and human oversight was built directly into the application stack, Brakebill noted. The result is a cross-functional feedback loop in which frontline associates validate outputs, improve model accuracy and take on higher-value work.

“I think that is the secret sauce — having a person in the middle of this who can use their judgment and reasoning to get to the right outcome and make the model better,” Brakebill said. “The consistency we’re seeing in our output is increased because humans have that judgment. We’ve seen more time for the team to focus on solving other problems.”

The speed of delivery has also shifted. Within 90 days, the team moved six or seven concepts to prototype-quality builds — a cadence that can be attributed to tight coordination across business, technology and implementation partners, according to Ramdoss.

“Responsible AI is essential, specifically in healthcare,” Ramdoss said. “Human in the loop is not an add-on process, but it needs to be part of the design.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of Google Cloud Next:

(* Disclosure: Deloitte sponsored this segment of theCUBE. Neither Deloitte nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE