Voice AI agents are one of the most discussed use cases for agentic AI in customer service. 

Billed as being able to hold natural-sounding conversations, voice AI agents can collect information from customers, respond and take limited, guardrailed action during customer interactions.

Mutual of Omaha Mortgage put both their customers and employees into the virtual hands of voice AI agents back in February 2025. 

The company deployed voice AI agents that automate the first 30 seconds of a customer interaction by gathering needed information and then passing the call to a loan officer. This shifted loan officers’ workflows from primarily outbound, manual dialing and then waiting for a prospect to pick up, to answering inbound calls.

Loan officers are one the company’s most valuable assets, Kevin Griffith, VP of sales and technology and intelligent platforms at Mutual of Omaha Mortgage, said Thursday at the Regal Rise virtual conference.

“Our loan officers’ skill is best when they have an actual opportunity on the phone. Dialing and waiting for a voicemail or a hello is not their strength,” Griffith said. “Voice AI allowed us to maximize their opportunities as it related to getting and having a conversation instead of spending hours and hours just waiting for the phone to ring.”

Getting to that point, however, involved wading through and evaluating dozens of voice AI vendors and their products. Griffith even built his own voice AI agent just so he’d know what “tough questions” he should ask when evaluating those vendors’ products.

And, being in a heavily regulated industry meant that Mutual of Omaha had to place extraordinary emphasis on guardrailing the voice AI agents so they didn’t do or say something that might damage the company by violating regulations or company policies.

Overcoming AI antipathy 

Because Griffith started as a loan originator, he knew what borrowers typically ask within the first 30 seconds of a call. That experience factored into the prompts used to build the voice agents.

“In such a regulated space, there are very clear things that you cannot say. You have to be able to exit out of those conversations. Drift is real. Hallucinations are real,” Griffith said. “I tell everybody, voice AI or just AI in general, is one of those things that it's easy to get it to say something. It gets a little more difficult getting it not to say something.”

But the hurdles weren’t merely technological. Griffith had to sell the users — the loan officers — on the forthcoming solution.

“Coming from a sales background, I knew there weren’t many loan officers who would ever say that an AI could outsell them,” he said. 

Beyond that, there’s generalized antipathy toward AI solutions since they’re often pitched as digital labor replacements for jobs people do. So, Griffith made it clear that wasn’t the goal.

“What we’re doing is starting the initial conversation of customer engagement. Very few questions up front, and we're getting that to you as a licensed banker, so you can do what's best,” he said. “Once they got over that hurdle, it was kind of like, OK, this is much more an opportunity than a concern.”

How the customers reacted to the solution was another hurdle. 

“You have some borrowers and the first thing they say is, ‘Are you AI?’ Not one word was said yet, and with full disclosures and everything else already in place, the immediate reaction was just ‘click.’”

Those abrupt hang-ups are largely driven by the experiences customers have had with legacy AI and interactive voice response systems and needing to repeat themselves when the speech system didn’t comprehend what they said.

“People have that in their minds already, and they think this is just another version of it. It just has a new name to it,” Griffith said, noting further, that the more “natural the AI sounds, the more conversational the AI sounds, is what we need to account for.”

Mutual of Omaha is doing that, in part, by measuring the sentiment of the calls — that is, how the borrower reacts to the voice AI agent.

Much of the benchmarking, sentiment and otherwise, relates back to the pre-voice AI environment. That involves answering two questions: Are customers happier and more content because they had the AI conversation up front? And is that initial engagement enabling the loan officers to better sustain clients and provide them with better service?

“The biggest risk we took was having AI handle that first 30 seconds. But we knew that in this space, the biggest impact would be in keeping our loan officers on the phone with a borrower,” Griffith said. “We are still measuring what we are kind of defining as success, but it’s kind of panned out that way.