XMarts room

Winter 2025

careCycle

Voice AI teams for medicare agencies.

Visit website

What they do

teams of voice AI agents for Medicare agencies

How close to us

5, very close

multi-agent handovers sold to one agency type; reports conversion and retention metrics; no loop learning stated

Wedges

  • W5 One layer for one type of small firm direct
  • W2 Seeing whether your AI agents actually produce adjacent
  • W3 Company memory from every call, for consultancies adjacent
  • W7 Find and fix margin leaks adjacent
  • W8 The other 90% of an AI skill adjacent
L1 Company brain (knows the organisation) 2, clear
L2 Agent team (runs agents, tools, approvals) 2, clear
L3 Does and learns (finishes, checks, measures, updates the brain) 2, clear

Who buys

Firm size
sme
Buyer type
business
Role
owner of a Medicare or ACA agency, call centre or FMO
UK
no
Their own description (from YC)

careCycle is an AI workforce for medicare agencies and FMOs - helping leading agencies achieve 5x higher conversions, 37% better retention, and saving thousands of agent hours.

We're transforming how Medicare agencies handle member relationships with teams of voice AI agents. We're solving the industry's biggest challenge: 50% annual member churn caused by poor post-enrollment engagement.

Our multi-agent AI teams work together to handle the entire customer journey - from initial contact through enrollment and ongoing care. When a lead comes in after hours, one AI agent answers and schedules a callback, another calls at the perfect time, a third captures qualifying information, and a fourth handles the warm transfer to a licensed agent. This approach helped a leading FMO write over $2M in renewals during AEP 2024 while reducing agent talk time by 20% per sale.

We've pioneered a purpose-built solution for Medicare's unique compliance requirements. Unlike generic AI platforms, our system features built-in HIPAA/CMS compliance, real-time data structuring, and seamless CRM integration. This enables agencies of all sizes to deliver enterprise-grade member care without massive infrastructure investments.

Early metrics show the power of our approach:

- 5x increase in after-hours conversion rates
- 37% improvement in 90-day retention
- 3,000+ agent hours saved in a single AEP
- $2M+ in additional renewals during AEP 2024

The Medicare Advantage market represents a massive opportunity, with over $360B in annual premiums and growing 10% year over year. By solving the member retention challenge, we're positioning ourselves to capture significant value by transforming the member care experience for beneficiaries.

Full teardown

careCycle (YC Winter 2025): teardown vs XMarts

In one line: careCycle sells an "agency operating system" (a CRM, a phone dialer and voice AI "employees") to Medicare and ACA insurance agencies, call centres and FMOs in the US [site] [yc]. Threat: 1, because it has our W5 shape (one layer for one type of small firm) and touches all three layers, but it sits in a US-only niche we are not in [site] (inference). It matters most as a model of what a vertical full-loop company looks like at 5 people.

1. Fact file

Item Finding Source
What they say Home page: "The agency operating system that compounds your book for you." YC line: "Voice AI teams for medicare agencies." [site] [yc]
Founded, team, funding Founded 2024, YC W25, 5 people, San Francisco. They launched in Q4 2024. A $2.0M seed round is reported for 11 Mar 2025, with no investors named on the page we could read. [yc] [press]
Founders Alex Doonanco (CEO): led go-to-market at AI Rudder, a voice AI firm, and says he helped it grow from $0 to $1M ARR in six months. Evan Roubekas (CTO): software engineering graduate who works on AI orchestration for Medicare. What YC likely saw: a founder who sold voice AI at scale plus one narrow regulated niche (inference). [yc] [press]
Customers No named logos. One unnamed "leading FMO" (AEP 2024), one unnamed independent agent with 2,500 members, one unattributed testimonial from "a call center owner" with 20+ agents. Integration partners named: MedicarePRO, GoGuruX, EnrollHere, Salesforce, HealthSherpa, Spark. [yc] [site] [blog] [press]
Price Public model, no dollar figures: "No tiers, no bands" flat rate per active policy, usage billed separately, $0 per seat. Their call-centre page describes buyers spending $150K to $200K a month on outbound. Job ads: none found (YC page shows 0 open roles). [site] [yc]
Path Q4 2024 after-hours voice AI for Medicare leads and welcome calls. Feb 2026 interview: began as an after-hours tool, grew into a broader engagement platform. Apr 2026 onward: weekly changelog of CRM, dialer and softphone features. 16 Oct 2026: "careCycle 2.0", a rebuilt app, with "Assembly" for the AI conversations. Former name Nodable Labs appears only in the YC list data; no other source found. [press] [site] [docs] [yc]

2. Their proposition, reverse engineered

Part What they do
Who buys Owners of Medicare and ACA agencies: independent agents, field agencies, call centres (page says 15+ agents), FMOs [site]. Small to mid-sized, US only [site] (inference on size).
The pain Members leave: they claim about 50% annual churn from weak care after the sale [yc]. Agents chase new sales instead of keeping members. Inbound calls go unanswered after hours. Agencies "ran everything after it on memory and spreadsheets" (their About page, paraphrased) [site].
The promise Keep and grow the book: more enrolled members, fewer lost ones, less cost per sale. They claim +15% effectuation and -20% cost per acquisition [site].
How it works AI employees ("Cora", now shown as "Assembly" in campaigns) take calls, qualify leads, book appointments, do welcome and check-in calls, and warm-transfer to a licensed agent [site] [yc]. Answers that need judgment go to a human for approval before delivery [yc] [site]. A CRM holds one record per member. Import a book from CSV [site].
The output A booked appointment or a handoff with context; an updated member record; enrolment data synced back from HealthSherpa; source tracking from lead source to sale [site].
Proof style Their claims: 5x after-hours conversion, 37% better 90-day retention, $2M+ renewals written by one FMO in AEP 2024, 3,000 to 4,000+ agent hours saved, 15% higher effectuation, 20% lower cost per acquisition [yc] [site]. Call counts disagree: "400,000 conversations" (Aug 2025, Feb 2026) vs "4M+ AI-handled calls" (site). In 2.0 they claim repeat notes fell from 14 to 1 and false callback promises from 33 to 6 in internal tests [docs].
Who sells it Founder-led sales plus conferences (title sponsor of Medicarians Vegas, Apr 2026) and CRM partner integrations [blog]. Asks for intros to agency owners in the launch post [yc]. Head of Growth and Head of Customer Experience are on the About page [site].
Price model Per active policy plus usage; free seats [site]. Revenue is described as tied to retention and renewals [press].
The story Their enemy: generic AI that fails in regulated work, and an agency that stops caring after the sale. Their belief: purpose-built for Medicare, "helpfulness over humanness", compliance first [press] [yc].

3. Positioning

  • Category they claim: "agency operating system for Medicare and ACA distribution" [docs]. Earlier: "AI workforce for medicare agencies" [yc].
  • Category a YC partner would put them in: vertical software with voice AI, a CRM-plus-AI play for one regulated niche (inference).
  • What they want to be known for: the one system a Medicare agency runs its whole book on, with AI doing the calls.
  • Where they sit on the map: layers L1/L2/L3 = 2/2/2, buyer = sme, business.

4. Three layers vs ours

Layer Them (score + evidence) Us (Marton's logs show)
L1 Company brain 2. A member record (policies, appointments, open requests, recent calls) that the AI and humans both read, plus saved scripts, tags and suppression lists [docs] [site]. Memory is about members, not the agency's own rules or decisions, so this is a borderline 2 (inference). A vault of 1,396 pages, 112 call transcripts and 8,763 customer quotes, searchable by meaning.
L2 Agent orchestration 2. Several AI roles with handovers: pre-screen, schedule, call back, warm transfer, with human approval for judgment answers [yc] [site]. Fixed per campaign; no sign of a general agent router. 5,641 jobs run by agents from three makers (Codex 4,756, Claude 839, Gemini 46), each checked by a different AI.
L3 Execution and learning 2. The agents do the work (calls, bookings) and results are measured: source tracking to a sale, policy status, retention [site]. Improvement is driven by staff: they review calls, prepare changes, test them in simulation and apply a new version [docs]. No automatic learning stated. About 160 incident records, each with an automatic check. Offer feedback lines fell from 44 to 18 with 22 lint rules added. Only about 15% of jobs reached main, and merges were unreliable in October.

5. Same and different

Topic Same or different Them Us
Shape of bet Same One vertical, all three layers, sold to a small firm that cannot build it [site]. W5 in the brief: one layer for one type of small firm.
Start small, then widen Same Began with after-hours calls and one number (5x after-hours conversion), then added CRM, dialer, revenue plays over about 18 months [yc] [site]. The brief's thesis: one workflow, one number.
Buyer Different Medicare and ACA agency owners, US [site]. SMEs and scale-ups, UK-leaning; no customer yet (brief).
Kind of work Different Phone calls with seniors, compliance-heavy [press]. Knowledge and ops work; coding, content, offers on Marton's own business.
Who owns the result Same They run the work and show the number (retention, renewals) [site]. Marton's logs show agents do and check work on his own business only.
How it improves Different Humans review and apply changes in Assembly, with versions [docs]. Marton's logs show rules added from failures and feedback by the system itself, with Marton approving.
Proof Different Vendor claims with no named customer; two AEP seasons live [site]. No outside customer pays.
Delivery Different Software, self-serve CSV import, "live in days" [site]. The system plus two founders, no productised version yet.
Price Different Flat fee per active policy, $0 seats [site]. Not set.

6. Wedge overlap

Wedge Fit Why
W5 One layer for one type of small firm direct Same shape: a vertical operating layer for a small regulated firm type [site]. Direct on shape, not on market.
W7 Margin leaks for commercial leaders adjacent Their Revenue Suite runs cross-sell and utilisation outreach on the existing book [site]. Different buyer and money lever.
W8 The other 90% of an AI skill adjacent Assembly gives versioned agent configs, simulation before going live and review of call performance [docs]. Human-driven, no auto upkeep.
W3 Company memory from every call adjacent They record, transcribe and keep a member fact sheet per call [site] [docs]. Different buyer.
W2 Seeing whether your agents produce adjacent In-product call review and a data assistant for their own agents only [docs] [site].

7. Whitespace (where they stop)

  • Layer gap: learning is by staff, not by the system. Assembly lets people review, simulate and apply changes [docs]. To own it we need a rule-change log tied to checked runs, with rework falling.
  • Buyer gap: only US Medicare and ACA distribution [site]. Nothing for UK firms or other trades. To own it we need one named buyer type with the pain on file.
  • Number gap: their numbers are claims with no named customer, and call counts conflict (400,000 vs 4M) [press] [site]. To own it we need one before/after number a customer would sign.
  • (Price and speed are not gaps: they publish a per-policy model and say "live in days" [site].)

8. For the YC application

  • Proven today: Marton's logs show his system runs agents that do and check work: 5,641 jobs across three AI makers, each checked by a different AI. About 160 failure records each added an automatic check, and offer feedback lines fell from 44 to 18 as 22 rules were added. Against careCycle: they show a vertical full loop is real, but their improvement step is staffed by people [docs].
  • Draft answer to "how are you different from careCycle?": DRAFT. "careCycle runs Medicare agencies in the US; we are not in that market. Their staff review calls and apply changes by hand. Our system changes its own rules from checked results." Founders say it their own way. (Keep the last sentence honest: it is proven on Marton's own work only.)
  • Future claim (needs proof): "One type of UK small firm runs on our layer, and its number moves." Needs one customer workflow run for 4+ weeks with a before/after number.
  • What we understand that they don't: a layer that learns from checked results on its own, not through staff review. needs proof on a customer; the only evidence is Marton's own offer-feedback drop.
  • Copyability (M3): medium. They already hold the scaffold: call data, versioned configs and simulation [docs]. Adding automatic learning is months of work and fits their buyer. But in a compliance-heavy niche they may keep a human sign-off on every change (inference). Not hard, so M3 can not score above 3 against careCycle. They would not enter our market (US Medicare focus, "single heavily regulated niche" [press]), so copying is a threat only if we pick Medicare.
  • Why now (M4): None found. Their dated change (CMS 2027 final rule, Apr 2026 [blog]) is a Medicare rule change that helps them, not us.
  • Do not claim: "most rivals do one layer" without a caveat: careCycle covers all three at level 2 in its niche. "Our agents learn automatically for customers." "We serve Medicare or any regulated phone work." Any careCycle number as fact.

M2 (what YC funds now) is not judged here. Theme labels above feed Pass 3.

9. Ideas worth taking

  1. Price on the thing the buyer already counts (active policies), no seat fees, no tiers. A small buyer understands it in one line [site].
  2. Simulate a change before it goes live, and keep a version history of who changed what [docs]. This is a clean answer to "how do you make learning safe?"
  3. Win one workflow with one number first (after-hours conversion), then widen to the whole book [yc] [site].

10. Unknowns

  • Revenue and paying customers. A search summary of a Techleap item (page blocked, 403) says about $1M ARR and one million conversations in Aug 2025 and names two investors: UNVERIFIED.
  • Named customers. Look for case studies, the docs site (docs.carecycle.ai) or conference talks.
  • Whether 4M+ calls or 400,000 conversations is right, and what counts as a "call".
  • Why YC lists "Nodable Labs". Ask or check Companies House-style filings for a US entity.
  • Whether any learning is automatic. Read the Cora and Assembly pages in full and the docs site.
  • Their open roles (the Work at a Startup page returned no readable list).

Sources

Technical detail
Slug
carecycle
Lens hash
179096c479
Pass
deep
Stale
no