Every year, many seniors in the United States who are already on Medicare have a two-month window for open enrollment during which they can make changes to their healthcare coverage. EHealth, Inc., a nationwide private health insurance agency, uses Regal’s Alice, an AI voice agent, to prescreen Medicare beneficiaries when they contact eHealth’s call center.
“That Medicare enrollment period is about 54 days long,” said Atul Kumar, VP of product and AI at eHealth, at the Regal Rise conference last month.
“We go through maybe half a million calls in those 54 days, and we work 14 to 16-hour days, including weekends, and we still can't get through all of them,” he continued.
Despite offering a website through which customers can enroll, 70% to 80% of eHealth customers buy through the telephone channel, Kumar said. This led to people waiting for six hours just to speak with a licensed human agent. One of eHealth’s goals was to reduce that wait time — and do so with a solution that was professional and empathetic.
Today, eHealth’s Alice answers every call immediately, has received 77% “exceptional” satisfaction ratings from callers and a 27% higher purchase rate than eHealth had when it employed human screeners via an outsourced call center, according to data Regal provided to CX Dive.
From pilot to full-time voice AI agent
EHealth’s voice AI deployment began in February 2025 with a pilot program. It offered an after-hours AI agent to handle calls that came in when the eHealth call center was closed. The learnings from that pilot’s performance informed the creation of the full-time voice AI agent.
Getting to that full-time screener required a great deal of iteration and problem solving.
Choosing a Medicare plan is complicated, emotional and personalized. Kumar related an anecdote of one caller who’d just lost her husband but didn’t explicitly say that. Instead, she said, “I have the remains of my husband with me.”
Alice understood what the caller meant and responded empathetically, he said.
Legacy versions of voice-enabled automation require explicit programming to recognize specific words.
But because Alice is LLM-based, it can understand what people are saying without having to be programmed for every conceivable variation of how someone might say their loved one had passed away.
“One of the gifts of AI in these scenarios is just consistently behaving the way you're supposed to behave, which is being empathetic, being customer centric and always being respectful,” Kumar said. “If you ask seniors what is the one thing they hate about call centers, it’s that people are not patient and they are disrespectful.”
Getting voice AI to play by the rules
EHealth must also comply with insurance and healthcare regulations. “There are very strict rules about what you can and cannot say or do in those conversations,” Kumar said.
One such rule requires the uninterrupted reading of a lengthy privacy statement. Often, callers would interrupt the reading, which would then require the automation, or the human, to start over.
“Nobody wants to hear those lengthy statements,” Kumar said. “They want to get to the ‘I want I want to talk to somebody’ and ‘Can you help me get a plan?’”
With Regal’s help, eHealth built a voice AI system that could tolerate the interruption.
“We developed the concept of a static node where the AI agent could say a specific statement, [that wasn’t] LLM driven, and not get interrupted,” said Regal co-founder and CTO Rebecca Greene, at the event.
The question then became how to re-engage the customer after they listened to the statement and transfer them back to the human agent for the actual conversation about Medicare enrollment and which plan might best suit their individual requirements.
That solution was simple: The voice AI agent asked, “Are you still there?” Regardless of how the caller responded, they were transferred back to the human agent.
There are additional complexities within self-enrollment, Greene noted. Some customers call on their own behalf, while others call with power of attorney, so there can be two people speaking on the line, and two, potentially, interacting with the voice AI agent.
“We didn't start out with the perfect agent from scratch,” Greene said. Getting to an increasingly robust voice AI agent took work.
“For any business that's regulated it's not a plug-and-play solution,” Kumar said. “There's no book written on solving these problems. There are no existing patterns to figure out how you do transitions and how you get around real-life problems when people interrupt you.”
For eHealth, it required trial and error and A/B testing — as well as qualitative measures like simply listening to a test call to make sure it sounded right even when the quantitative metrics said the agent was correctly functioning.
EHealth had an all-hands-on-deck philosophy when it came to handling the open enrollment period of peak call volumes. As a result, the team Kumar assembled to tailor Alice to the company’s needs had firsthand knowledge of what a good call should sound like.
“That was a pretty cool and unique experience that I think allowed the agent to get better really quickly,” Greene said.
Overcoming stakeholder concerns
Kumar noted that while stakeholders within eHealth saw the opportunity for voice AI to streamline their inbound calls, there was quite a bit of fear, as well.
“What if AI says something that is completely outrageous [or] hallucinates?” Kumar said. “One thing we did was have everybody involved, so we had compliance as a key stakeholder working with us.”
To allay those fears and develop buy-in, Kumar’s team included those compliance stakeholders from other departments — from quality assurance and telephony engineers, to CX managers and trainers. Kumar also played some of the interactions Alice had with customers to senior leaders so they, too, could understand what the AI agent was saying and doing.
“Finding people to work with wasn’t difficult. Everybody wanted to be part of an AI initiative in the company,” Kumar said. “But just getting it right and making sure you keep focus on what matters the most, which is the customer experience and, of course, the business KPIs.”
