Product brief & business plan
Clutch is an AI concierge that makes the calls, waits on the hold, and runs the errands you dread — from one plain-English request.
Executive summary
Clutch is a consumer AI concierge that completes the phone calls and life-admin tasks people avoid — booking, cancelling, negotiating, and above all waiting on hold — from a single plain-English request.
The product is live in production with eight task types; its flagship real-call engine is built, reviewed, and one supervised call from launch. The market for “AI that makes calls for you” arrived in 2026 — and that is the opportunity, not the threat. A wave of single-purpose call agents has proven consumers want this; none of them tie the calls to the full sweep of life-admin, package the work as outcomes rather than a phone tool, or lead with trust. Clutch competes on breadth, concierge-grade outcomes, and a disclosure-first stance, monetized as a simple $15/month upgrade over a free tier.
The problem
Booking, cancelling, disputing, waiting on hold, chasing a quote, getting a sooner appointment — the life-admin tax. It’s not hard. It’s just tedious, synchronous, and stuck behind hold music. So we all put it off, pay the late fee, keep the plan we meant to cancel, and take the appointment three weeks out.
Figures above are illustrative framing of a universal, well-documented friction — not audited market data.
What Clutch is
Type get me an earlier dentist appointment
or sit on hold with the airline and put me through to a human.
Clutch reads the intent, picks the right kind of task, fills in the details it can, and goes to work — placing the calls, navigating the menus, negotiating, or watching for an opening.
You get a live view of what’s happening, an itemized receipt when it’s done, and a decision only when one’s actually yours to make. It’s the difference between a to-do list and a concierge desk.
The catalog · nine ways to hand it off
Clutch dials a support line, works through the menu, waits on hold, and rings you the moment a real person picks up. Wait on hold with Comcast and get me a human.
Continuously watches for an earlier slot — appointment, reservation, opening — and grabs it the instant it appears. Find me a sooner opening.
Books the table — and will call the restaurant to cancel it later if plans change. Table for four Friday at 7.
Books doctor, dentist, DMV, salon, or vet visits around your availability. Schedule a checkup with Dr. Lee.
Calls the provider and works the bill down, then reports what it saved. Lower my internet bill.
Ends the subscription or service you keep forgetting to kill. Cancel my gym membership.
Gathers and compares prices so you don’t make five calls yourself. Get me quotes on movers.
Gets you on the list and watches for your name to come up. Get on the standby list.
Anything that doesn’t fit a lane — described in your own words. Whatever you need.
The flagship
This is the part that isn’t a demo. Holdline places a genuine outbound phone call, hears the line in real time, talks its way through the menu, and waits — so you don’t sit there with the phone to your ear. It’s built on the same voice + reasoning stack the biggest AI labs ship, wired into a purpose-built state machine.
Places a real outbound call to the number you name.
Listens to the IVR and speaks or presses its way to a human.
Waits out the hold music — for as long as it takes.
Recognizes a real person, and says up front that it’s an AI.
Rings your phone and connects you the instant the human’s there.
Under the hood: Twilio ConversationRelay for the live voice leg, Claude (Opus) for menu navigation and human detection, and a purpose-built, fully unit-tested call state machine with a human-detection guard so hold announcements can’t trigger a false handoff.
Traction · where it stands today
Clutch is live in production with a full account system and eight task types running today. The live-call engine is built, reviewed, deployed, and dark — one real-world validation pass away from switching on. This is a side project that has quietly become a real product.
Deployed at clutch.mosaic.site on managed Postgres + object storage. Accounts, login, password reset, memory that pre-fills your next request, itemized receipts, and shareable recaps — all in production.
Reservation, Appointment, Bill, Cancel, Quotes, Waitlist, Sniper, and Errand run today, each with guardrails against split-brain state, cross-user access, and stranded jobs.
Real Twilio + Claude call path, built and code-reviewed, deployed behind a feature flag. Ingress for the live voice socket is confirmed in production. One supervised real call away from launch.
50+ automated tests — unit and full end-to-end integration suites — passing, with a self-cleaning harness that leaves production untouched. Built test-first, task by task, each reviewed before merge.
Market analysis
Every phone-owning adult is a potential user — the tasks Clutch handles are ones nearly everyone faces. The reachable market is the slice who will pay to never do them again: busy households, professionals, and anyone for whom time beats the phone. The tiers below are an illustrative top-down model with assumptions stated — framing for scale, not audited research.
U.S. adults who regularly deal with bills, bookings, cancellations, and hold queues.
Digitally-comfortable adults open to an app placing calls for them — the behavior 2026’s call-agents have now validated.
~1% of SAM in the first phase. At $15/mo, that is already a meaningful, self-funding base.
Illustrative model — TAM ≈ U.S. adult population; SAM ≈ share comfortable delegating calls to software; SOM ≈ an early-penetration wedge. Conversion and pricing are proposed, not observed. Intended to frame opportunity size, not forecast revenue.
Competitive analysis
“AI that waits on hold” went mainstream in 2026 — proof of demand, and proof we’re not alone. But today’s players are point tools: a call agent, or a bill-fighter, or a booking bot. Clutch’s bet is that people want one place that does the whole sweep, reports the outcome, and is honest about being an AI.
| Player | Real end-to-end calls | Breadth of tasks | Cost structure | Trust & transparency | Consumer price |
|---|---|---|---|---|---|
| ClutchAI concierge · 9 task types | ●Dials, navigates, holds, bridges | ●Broad — calls + bookings + negotiation + more | ●AI — cents per task | ●Discloses AI, never records, receipts | Free / $15 mo |
| AI call agentsGetHuman, Assindo, Simple AI, HeyRobyn | ●Yes — the category we share | ◐Narrow — calling only | ●AI — low | ◐Varies by app | Free–$/mo |
| AI consumer-rightsDoNotPay | ◐Mostly chat / form-based | ◐Disputes, bills, cancellations | ●AI — low | ○FTC settlement (2025) over AI claims; low trust scores | ~$36 / 2 mo |
| Human assistantsFancy Hands, EAs, task services | ●Yes — a person calls | ●Broad — anything | ○Human — $/task, doesn’t scale | ●High — it’s a person | $$–$$$ / mo |
| Do it yourselfthe status quo | ●You, on hold | ●Anything — if you find the time | ○Your hours | — | “Free” |
The only option that is broad and cheap and honest: AI-cost economics like the call-agents, task breadth like a human assistant, and a disclosure-first stance that is the deliberate opposite of the category’s trust problems. Outcomes — receipts, a “won,” a decision only when it’s yours — turn a phone tool into a concierge.
Hold-waiting is no longer novel, and some rivals are free or funded. Breadth is our edge but not yet a moat; the defensibility we’re building is task-completion data — IVR trees, negotiation patterns — and a trusted consumer brand. We plan against incumbents adding breadth, not around the hope that they won’t.
The business
Adoption comes from letting anyone hand off a task for free. Revenue comes from the people for whom that becomes a habit — the ones who want the calls made, the bills fought, and the hold-waiting done without thinking about it.
Pricing shown is the proposed launch model and is adjustable. Unit-economics notes are illustrative of the structural advantage of an AI-run concierge versus a human one — not a financial forecast.
Go-to-market
Acquisition starts where the hurt is sharpest and the demo most visceral — waiting on hold — then widens into the full concierge once trust is earned. The best marketing is a result worth screenshotting.
Wedge
Expand
Compound
Financial outlook
The structure is favorable: each AI call costs cents in voice and model minutes, while the value it replaces — an evening on hold, a human assistant — is worth orders of magnitude more. The three-year sketch below shows shape and operating leverage on the proposed $15/mo price and stated assumptions. It is an illustration, not a forecast or a commitment.
| Illustrative model | Year 1 | Year 2 | Year 3 |
|---|---|---|---|
| Free users (cumulative) | 25,000 | 150,000 | 500,000 |
| Paid conversion | 4% | 5% | 6% |
| Clutch+ subscribers | 1,000 | 7,500 | 30,000 |
| Annual recurring revenue | ~$180K | ~$1.35M | ~$5.4M |
Illustrative only — ARR = subscribers × $15 × 12. User growth, conversion, and price are assumptions chosen to show operating leverage, not projections of actual results. Marginal cost per completed call is expected in the single-digit-cents range, well under the subscription’s per-task value.
Team & operations
Clutch was designed and built end-to-end by its founder — the full product, the account system, and a real telephony + AI call stack with a fully-tested state machine and safety guardrails. What a solo build has already shipped is the strongest signal here; what it needs next is deliberate resourcing.
Solo founder-operator. Live product, eight task types, real-call engine in pre-flight, 50+ automated tests, self-cleaning infrastructure. Runs on managed cloud (Postgres, storage, containers) — low fixed overhead.
Reliability & on-call coverage for live calls, a growth/GTM partner, and support as volume grows. The engineering foundation is built to widen the allow-list — not to be rebuilt.
Risks & mitigations
Funded or free competitors
AI call-agents already exist; some are free or well-backed.
Compete on breadth + outcomes + trust, not “can do calls”; build the task-completion data advantage early.
Trust & regulation of AI calls
AI placing calls raises disclosure, consent, and recording-law questions — and rivals have been burned.
Disclosure-first by design: always announces it’s an AI, never records, hard caps. Trust as a feature, not an afterthought.
Real-world call reliability
IVRs, accents, and edge cases are messy, and the live engine is pre-flight.
Ships dark behind a flag + tight allow-list; a supervised first call, then widen deliberately — human-detection guard already in place.
Unit economics at scale
Voice + AI minutes are cheap but not zero; heavy free use could add up.
Free tier is metered; the expensive live-call tasks sit behind Clutch+, so cost scales with paid usage.
Single-founder concentration
Key-person risk on a solo build.
Documented, tested, handoff-ready codebase; resourcing the team is the explicit next step.
Platform dependency
Relies on third-party telephony and model providers.
Provider-agnostic call and AI seams; the state machine and product logic are ours and portable.
Where it goes next
Now — pre-flight
Next — the wedge
Then — the vision
The short version
Clutch takes the most universally hated part of adult life — the phone calls and the waiting — and quietly does it for you. The product is live, the hard engineering is done, and the flagship is one real call from launch. What’s left isn’t whether it works. It’s how far it goes.
Try it → clutch.mosaic.site