Case study: invoice generator MVP shipped in 6 days

Founder reached out on a Sunday. By the next Sunday there were three paying customers. Here is the actual timeline, with the parts I usually don't put in marketing copy.

The brief

A freelance accountant who wanted to sell a stripped-down invoice generator to her small-business clients. Send a Telegram message with the client name and amount, get a PDF invoice back, billed monthly per active client.

The riskiest assumption: would her existing client base actually pay for this when they could use a free template? That was the only thing the MVP needed to answer.

Day 1 (Monday) — scope and design

3-hour kickoff call. We cut:

• Multi-user accounts (she was the only user)
• A web dashboard (Telegram was the surface)
• Logo upload (hard-coded for v1)
• Multiple currencies (USD only)
• Tax line items (one tax rate, set once)

What stayed: a Telegram bot, a PDF generator, a Stripe link, a tiny admin page for her to see active customers.

Day 2 — build the spine

Telegram bot with grammY. Three commands: /new, /list, /help. Wrote the conversation flow first, then the storage. By end of day, the bot accepted a message and stored it in Supabase. No PDF yet.

Day 3 — the PDF

This is where I always underestimate. PDF generation is fiddly. Spent 5 hours getting a clean A4 layout in pdf-lib that worked across viewers. The trick: don't try to be clever with layouts. One column, generous margins, system font. Done.

Day 4 — the wall

Cloudflare Workers don't ship with the Node fs module. pdf-lib needed font files. I had wired font loading to read from disk, which works locally and explodes on deploy.

Fix: inline the font as a base64 string at build time. Cost me 6 hours and one emergency Discord question. Lesson re-learned: test deploys on day 2, not day 4.

Day 5 — Stripe and the admin page

Stripe payment link (not full Checkout — too much for a 6-day budget). A 50-line Next-ish admin page so the founder could see who paid and revoke access. Manually wired the webhook.

Day 6 — show, fix, ship

Showed the founder. She found 4 things I'd missed:

1. The PDF had her old business name
2. /new should remember the last client
3. Stripe link should pre-fill her email
4. The bot's welcome message was too long

All four were 15-minute fixes. Shipped Friday evening. First paying customer Saturday morning.

What I'd do differently

• Test the production deploy on day 2, not day 4.
• Cut the admin page entirely. She would have used Supabase Studio for a week and been fine.
• Charge slightly more. Fixed-price sprints look great until the day 4 wall.

The whole thing was 35 billable hours. Three months later it's at 40 paying customers with zero ongoing work from me. That's the bar for "good MVP outcome" — you ship something, it buys real information, and you're not on the hook to babysit it.