The Robot That Answers Medicare's Second Phone Call
careCycle put voice AI on the line for Medicare agencies - the pre-screens, callbacks, and after-hours questions a human team can never fully staff. Then it hands the licensed sale back to a person.
Every year, on January 1st, a strange thing happens across the Medicare business. Roughly half the members an agency signed up the previous fall quietly disappear. Not because the plan got worse. Not because a competitor undercut the price. Most of them leave because, somewhere in the long silence between enrollment and renewal, they called with a question and nobody picked up.
careCycle, a San Francisco company from Y Combinator's Winter 2025 batch, was built for that silence. It sells voice AI teams to Medicare agencies and the Field Marketing Organizations, or FMOs, that sit above them - software that answers the phone at nine at night, calls a member back at the hour they're most likely to pick up, screens a new lead before a human ever spends a minute on it, and then, when there's an actual licensed sale to close, warm-transfers the call to a person.
That last part matters, because it is also where careCycle draws its line. The AI does the work that does not require a license. The human does the work that does. "A helpful robot is better than an unhelpful human," founder and CEO Alex Doonanco has said, before adding the sharper half of the thought: "but there's a third option. People are used to an unhelpful robot." careCycle's whole bet is that seniors on a Medicare hotline will forgive a machine for being a machine, as long as it actually helps.
01 / THE LEAKA $15 billion silence
Medicare distribution is a bigger business than most people outside it realize. More than 40 million Americans are enrolled in Medicare Advantage or ACA plans, and the agencies that sell those plans are paid to keep members enrolled year after year. When a member churns, that recurring payment stops. Doonanco pegs the annual loss to member churn at around $15 billion - money that leaks out of the system not through fraud or competition, but through inattention.
The reason the leak persists is unglamorous. Selling a Medicare plan is a seasonal sprint - the Annual Enrollment Period runs from mid-October to early December, and agencies staff up, burn out, and then have to keep every one of those members happy for the other ten months of the year. During the crunch, calls go unanswered. During the quiet months, nobody has time to check in. The member who can't get a straight answer about a benefit is the member who switches plans the moment a competitor calls first.
Most of the pain lands on small operators. About 65% of Medicare brokers run one-person or tiny shops - no call center, no night shift, no IT department. When they're on a call, the next three callers go to voicemail. careCycle's argument is that these are exactly the people for whom an always-on AI team changes the math, because they were never going to hire their way out of the problem.
It helps to picture the member on the other end. A 72-year-old opens a letter about a formulary change and has one question about whether a prescription is still covered. If the call goes to voicemail, the anxiety doesn't. By the time the agent calls back two days later, a competitor's mailer has already arrived promising something simpler. The window in which a small question becomes a switched plan is short, and it usually opens outside business hours. That timing gap - not the quality of any single plan - is the market careCycle is really working in.
02 / THE PRODUCTNot a chatbot, a chain of agents
careCycle is careful not to describe itself as "an AI." It describes itself as a team - a set of coordinated voice agents, each doing one job in a sequence a human used to run alone. One agent picks up inbound calls around the clock. Another pre-screens and qualifies new leads. Another books appointments. Another runs a mid-cycle hotline for the coverage questions that arrive in February and June, long after the sale. Another places outbound calls at the times a given member tends to answer. And when a conversation reaches the point of an actual enrollment decision, the system warm-transfers to a licensed human agent.
The design choice underneath all of this is memory. careCycle's agents are built to hold context across calls - to recall that this member phoned last month about a dental benefit, to pick up the thread rather than start from zero. In an industry where the same senior often calls three times about the same question, Doonanco frames continuity, not a convincing human voice, as the real product. "We're not chasing humanlike AI," he has said. "We're chasing domain specificity and relationship continuity."
Splitting the journey into agents rather than shipping one all-purpose assistant is a practical decision as much as a technical one. Each agent has a narrow job, which means each one can be tuned, measured, and constrained on its own terms - the pre-screener optimized for accuracy, the hotline optimized for clear answers, the outbound caller optimized for reaching people at the right hour. It also mirrors how a Medicare office already thinks about its own work, where the front desk, the scheduler, and the licensed agent are different roles even when one person is filling all of them. careCycle is, in effect, selling the roles that go unfilled after five o'clock.
Alongside the voice agents, careCycle ships a CRM - call logging, member records, dialing - and does something counterintuitive with it: gives it away. Most Medicare software is priced per seat. careCycle hands the database over free and charges for the AI workforce that uses it. The effect is that the humans and the robots work off one shared member record instead of two disconnected systems, and the agency's pricing is tied to the thing it actually loses sleep over - retained members - rather than the number of logins.
03 / THE PROOFOne million conversations
careCycle launched in October 2024 and has been counting conversations ever since. By August 2025, when the company announced its funding, it had crossed one million conversations with Medicare beneficiaries, roughly doubling its volume over the prior month. The early results it points to are specific: a leading FMO that used careCycle wrote more than $2 million in renewals while saving over 3,000 agent hours during enrollment. One client saw after-hours conversion jump fivefold. Another improved 90-day member retention by 37%.
book
engagement
There is a smaller, quieter number that may say more about the product than the headline metrics. One independent agent, careCycle says, held onto a book of 2,500 members without hiring anyone. That is the kind of leverage that rarely makes it into a press release - not a workforce replaced, just a solo operator who got to keep the ground they would otherwise have lost. In a business full of one-person shops, staying in business is the whole game.
04 / THE DIFFERENCECompliance as the feature
The voice AI category is crowded. Horizontal platforms - the AI Rudders, Retells, Blands, Sierras, and Decagons of the world - can all, in principle, make a phone call. What they generally cannot do out of the box is know what a Medicare agent is legally forbidden from saying. Medicare marketing is governed by CMS rules, and member data is governed by HIPAA. An AI that improvises the wrong sentence on a recorded line is not a cute bug; it is a regulatory event.
careCycle's answer is to build those constraints into the agents themselves - compliance workflows, human-in-the-loop dashboards, and guardrails on what the AI is and isn't allowed to say. It's the difference between a general-purpose voice tool pointed at healthcare and a system designed from the first call for the one industry where the rules are the product. Doonanco's own background sharpens the point: he was the first U.S. hire and go-to-market lead at voice AI company AI Rudder, and he grew a horizontal voice AI business from zero to $1 million in ARR in six months before deciding the bigger opportunity was to go narrow.
Going narrow also changes what the company has to be good at. A horizontal platform has to serve a florist and a debt collector and a dental office with the same core product, which means it optimizes for flexibility. careCycle can do the opposite - it can bake in the specific vocabulary of Medicare Advantage, the rhythm of the enrollment calendar, and the exact phrases a compliance officer flags, because it only ever has to serve one kind of customer. The narrowness that looks like a limit from the outside is what lets the product be opinionated on the inside.
| Capability | careCycle | Horizontal voice AI |
|---|---|---|
| Built for Medicare / ACA | Yes | General purpose |
| HIPAA & CMS guardrails | Built in | DIY |
| Memory across member calls | Yes | Usually per-call |
| Warm transfer to licensed agent | Native | Bolt-on |
| Bundled CRM | Free | Separate |
05 / THE BUSINESSCharging for outcomes
careCycle's revenue comes two ways. Small agencies, on the order of five to ten employees, pay a subscription in the neighborhood of $10,000 a month. Larger relationships with FMOs tend to be outcome-aligned - tied to the member retention and renewal numbers careCycle can move. The company reported roughly $70,000 in revenue in its first month of operation and about $1 million in ARR around the time of its August 2025 raise, when it took in $2 million from Strike Capital and the Pioneer Fund, a YC alumni fund.
The seed round arrived with the CRM launch, and the two decisions are connected. Giving away the database while charging for the AI is a way of making the software indispensable before it is expensive - the member records live in careCycle, so the calls do too. It's a wedge that only works because the callers are the hard part and the database is the commodity, which is the reverse of how most vertical SaaS is priced. For a category as sleepy and phone-bound as Medicare distribution, that inversion is part of what makes the company worth watching.
Where careCycle fits in the market, then, is a narrow slot with a wide floor: the non-licensed engagement layer for an entire distribution channel that still runs on the telephone. It is not trying to be the Medicare plan, or the licensed agent, or the general voice AI platform. It is trying to be the thing that answers when none of them can - and to remember the conversation the next time the phone rings. Whether that's enough to hold back the January exodus is the question the next enrollment season will answer, one callback at a time.