careCycle Wants to Fix Medicare's Churn Problem With Voice AI Teams

The Macro: Medicare Is a Massive Market With a Retention Crisis

Medicare serves over 65 million Americans. The annual enrollment period generates a frenzy of activity every fall, with agencies competing to sign up members across Medicare Advantage, Medicare Supplement, and Part D plans. What happens after enrollment is where the system breaks down. Agencies sign people up and then largely ignore them until the next enrollment period. The result is roughly 50% annual member churn. Half of all enrolled members leave their plans or switch agencies every single year.

That churn rate is staggering when you look at the economics. Acquiring a Medicare member costs agencies between $300 and $800 depending on the channel. Losing half of them every year means agencies are stuck on a treadmill, spending enormous amounts on acquisition just to stay at the same member count. The agencies that solve retention gain a compounding advantage. Every retained member is one you don’t have to re-acquire.

The root cause is straightforward. Post-enrollment engagement requires regular, personalized outreach. Checking in after a new member’s first doctor visit. Explaining benefits they might not know about. Helping with plan changes when circumstances shift. Most agencies can’t staff this level of ongoing communication. They have enough people to handle enrollment season, and then those people move on to the next sales cycle. The members who signed up in October are effectively orphaned by January.

Voice AI is a natural fit here. Medicare members skew older. They prefer phone calls to apps and portals. The conversations are structured and somewhat predictable. Plan details, benefit explanations, provider networks, prescription coverage. This is exactly the type of interaction where voice AI can perform well because the domain is bounded and the stakes per individual call are manageable.

The Micro: AI Agents That Actually Handle the Phone

careCycle deploys multi-agent AI voice teams for Medicare agencies and Field Marketing Organizations (FMOs). Not a single chatbot. Multiple AI agents that can handle different parts of the customer journey. Inbound calls get an AI receptionist. Outbound campaigns run through AI-powered dialers. Member profiles live in a built-in CRM. The whole system is designed so an agency can automate the post-enrollment engagement that they currently don’t do at all.

The product is SOC 2 Type 2 and HIPAA compliant, which is table stakes for anything touching healthcare data but still worth noting because many AI voice companies in other verticals haven’t bothered with healthcare-grade compliance. The integrations connect to Salesforce, HubSpot, and AgencyBloc, the last being the CRM that a large portion of insurance agencies actually use.

Alex Doonanco is a founder who previously built a horizontal voice AI company from zero to $1 million ARR in six months. That’s a fast trajectory, and the fact that he chose to leave it and start something vertical in Medicare suggests he saw a bigger opportunity in going deep rather than staying broad. Evan Roubekas is a founder and software engineering grad from UVic who specializes in AI orchestration. They have a five-person team in San Mateo, part of YC’s Winter 2025 batch.

The “multi-agent” framing is interesting because most voice AI products are single-bot systems. careCycle’s approach is to have specialized agents for different functions. One handles inbound calls. One runs outbound campaigns. One manages scheduling. The coordination between these agents is what Roubekas’s orchestration background is presumably focused on. Getting multiple AI agents to share context about the same member across different interaction types is a harder problem than getting one bot to handle a single conversation.

Their reported numbers are aggressive. 5x higher conversions and 37% better retention compared to the status quo. Those are plausible when the comparison baseline is “agencies that do almost no post-enrollment outreach.” Going from zero engagement to automated regular contact should produce a large improvement. The more important question is how those numbers compare to agencies that actually have human engagement teams. Against a well-staffed retention operation, the AI advantage is probably smaller. But most agencies don’t have well-staffed retention operations, which is the whole point.

The Verdict

The market fit here is unusually clean. Medicare agencies have a quantifiable retention problem. The members they lose have a known acquisition cost. Voice is the preferred channel for the target demographic. And most agencies lack the staff to do the work manually. careCycle is selling automation for a process that agencies know they should be doing but aren’t. That’s a much easier sale than convincing someone to adopt AI for something they’re already handling with humans.

The pricing tiers scale from independent agents with up to 1,000 members to enterprise accounts above 5,000. That range covers the bulk of the agency market. FMOs, which oversee networks of independent agents, could be particularly high-value customers because a single FMO deal could mean rolling out careCycle across dozens of agencies.

The competitive concern is that voice AI for insurance is attracting a lot of attention. Levo.ai, Skit.ai, and other vertical voice AI companies are building for adjacent markets. If careCycle’s success becomes visible, expect larger voice AI platforms to add Medicare-specific features. The defensibility has to come from the compliance infrastructure, the AgencyBloc integration, and the domain-specific training data that accumulates as more agencies use the system.

At 30 days, I’d want to see call completion rates and member satisfaction scores. At 60 days, whether the retention improvement holds or whether members are just delaying their churn. At 90 days, the key metric is agency expansion. If agencies that start with one AI agent team add more, that’s the signal that the product delivers measurable ROI. The 4 million calls they report handling suggests they’re already past the proof-of-concept stage. Now it’s about proving the retention numbers hold at scale.