Project 21 finally clears $6K after 20 SaaS misses

The Case: $6,092 in 30 days from a 72-hour build

What he does. He ships an AI post-call workflow tool that turns a consultant's Zoom call into a branded recap email, an action-item list, and a rough proposal — before the coffee cools.

Backstory. By late 2025, AI note-takers collapsed into a commodity: Fireflies, Otter, Granola, everyone had one. But solo consultants and fractional operators don't need transcripts. They need deliverables. The gap between "here's what was said" and "here's what I'm sending the client tomorrow morning" was still 90 minutes of unpaid admin. Project 21 lives inside that 90 minutes.

Product. - Joins Zoom/Meet calls via Recall.ai - Generates a client-branded recap email in the consultant's voice within 4 minutes - Auto-drafts a follow-up SOW or proposal from the scope talk in the call - Pushes internal action items to Asana or ClickUp - One-click "send as-is" or quick edit in browser

Pricing. $79/mo per consultant, unlimited calls. Also a $199 one-time "meeting-to-proposal" starter pack for skeptics who won't subscribe cold.

Key numbers. - Total build time: 72 hours across 4 days - Time to first paying user: 9 days - Time to $1K MRR: 14 days - Month 1 revenue: $6,092 - Subscribers at day 30: 62 at $79/mo = $4,898 - One-time packs sold: 6 at $199 = $1,194 - Free trial → paid conversion: 41% - Cold-DM close rate inside IndieHackers/Slack: 1 in 27 - Refund rate: 1 in 62 subs (1.6%) - Direct API cost per subscriber/mo: $4.10 (94.8% gross margin)

Stack. Next.js 15, Supabase, Stripe, Claude 4.5 Sonnet, Recall.ai, Resend, Vercel, Loops.so for onboarding.

Why this works (and what most readers will miss)

1. The wedge is time, not tokens. He isn't selling "AI meeting notes" — a thousand tools do that for $10/mo. He's selling the 90 minutes between the call ending and the client email being sent. A consultant billing $250/hr pays $79/mo without doing the math. The commoditized layer is priced against; the un-commoditized layer is priced with.

2. The 20-failure discount. By project 21 he already knew which subreddits actually convert, which pricing pages don't need a demo video, which onboarding email drives the trial start, which Stripe webhook fails silently. That knowledge is invisible in the case and priceless in the outcome. A first-time builder would need six months to learn what he compressed into 72 hours.

3. The tool sells against a workflow, not a competitor. No comparison page. No feature matrix. The landing page just shows a real consultant's Monday morning inbox with four recaps sent before 9am. You're not comparing him to Fireflies. You're comparing him to your own Sunday-night dread. Nothing beats a mirror.

The Distilled Read

Sell the hour, not the model. Every AI-wrapper case in 2026 that's making real money has stopped competing at the transcription and summarization layer and moved one step downstream — to the deliverable, the invoice, the proposal, the email that ships to a client. Consultants don't buy AI. They buy the recovered evening. Price it accordingly. $79/mo for something the buyer would pay $250 to have done manually is a rounding error in their P&L and a 94.8% margin business for you. The math wins itself when you pick a buyer whose hour has a real price tag.

Two products from one engine. Look at what he did with the $199 one-time pack. Same infrastructure, same LLM call, different wrapper — a starter kit for the consultant who won't sign a subscription without proof. That single decision widened his top-of-funnel by roughly 40% without adding a line of code. The pattern travels: if you have a subscription, you have a one-time product. If you have a one-time product, you have a subscription. Ship both on day one and let the buyer self-select. Most operators leave the second SKU on the table for a year and then wonder why growth stays flat.

Margin math or die. $4.10 in API cost per subscriber per month means he can burn $50 in Meta ads to acquire a $79/mo subscriber and still hit a 3-month payback. Most 2026 AI wrappers are running 40% margins because they route every request through the biggest model when Haiku or Sonnet-mini would do the job. Pick your model per task, cache the boring calls, batch the async ones. The difference between 60% and 95% margins is not one skill — it's the difference between a business that can advertise and a business that can only pray for organic growth.

Show up where the pain already gathers. He didn't launch on Product Hunt. He didn't post a thread. He spent two weeks doing free audits inside three consulting-adjacent communities — a Slack, a Circle, and a specific subreddit — and closed 1 in 27 cold DMs into a trial. That isn't "marketing." That's being useful in the room where the problem already lives. The room existed before he arrived. His job was to walk in with the right tool and shut up until someone asked what he did for a living.

The moat nobody in the thread is naming. The comments are congratulating him on the product. The product is fine. The product is not the moat. The moat is his internal database of 20 dead SaaS launches — every one of which taught him which distribution channel was noisy, which pricing page killed conversion, which onboarding step dropped 30% of users, which vertical had cheap buyers who churn on month 2. Eighteen months from now, the winners in AI-wrapper land will not be the ones who wrote better prompts. They will be the ones with the longest failure ledger. He didn't succeed in spite of the 20 misses. He succeeded because of them. That's a compound asset you cannot buy with a YC check or a Stanford co-founder. And it's the one thing the 22-year-olds shipping their first project this weekend cannot copy — no matter how clean their landing page looks.

Steal-the-Playbook

1. Pick one vertical, not the category. Not "AI for professionals." Solo consultants who bill hourly and hate Sunday admin. That specific.
2. Find their existing room. One Slack, one subreddit, one Circle. Spend 14 days doing free audits and shipping recaps by hand. Log every objection you hear.
3. Ship the wrapper in 72 hours. Next.js + Supabase + Stripe + your best model. If it takes longer than 4 days, the scope is wrong. Cut features until it fits the week.
4. Price against the recovered hour. If your buyer bills $200/hr, price at 30-40% of one hour, monthly. $79 isn't a coincidence — it sits below the flinch threshold for anyone who bills.
5. Launch two SKUs the same day. One subscription, one one-time. Same backend, different landing page. Let the market vote before you build a roadmap.

Stack: Next.js + Supabase + Stripe + Claude API + Recall.ai. Under $80/mo infrastructure. 4 days to launch.