How long does it take to build an app? MeltsWell, month by month.

The honest answer to "how long does it take to build an app" is: it depends on how much you build before you show it to anyone. So instead of a range, here's what one real app took, with the git log as the source.

MeltsWell is a marketplace and leaderboard for home bakers and small sweet shops. Buyers find them, orders happen on WhatsApp. Web app, three languages, subscriptions, admin panel, blog. One person building, with AI doing a lot of the typing.

First commit: March 25, 2026. Public launch: early August 2026. Nineteen weeks. 1,723 commits.

Month by month

Month	Commits	What got built
March (1 week)	13	Waitlist page, Supabase, email, first deploy on Vercel
April	91	Seven-step baker onboarding, city search, admin review panel, language switching
May	150	Leaderboard, public profiles, creator dashboard, rebrand from the old name, pre-launch hardening
June	472	Stripe billing, pricing tiers, performance work, image optimisation, security fixes
July	455	Billing rewrite (free tier, new prices), brand pages, 28 blog posts, first six creators onboarded by hand
August	413	Public launch, Product Hunt, launch directories, mobile app started, account deletion
September (so far)	129	Coupons, onboarding nudges, French and Arabic fixes, mobile app screens

Two things stand out.

The first version existed by mid-May. Onboarding, profiles, leaderboard, admin. Seven weeks. That's the product. Everything after was billing, polish, content, and languages.

June and July were the expensive months. Nearly a thousand commits, and almost none of it was the core product. Billing alone was rewritten twice: the May spec had two paid tiers and a three-month trial, the July version added a free tier and cut prices in half. Both were right for their moment. Both cost weeks.

What took longer than expected

Billing. Stripe is easy. Deciding what to charge, what's free, what happens when someone stops paying, what to email them and when: that's the month. We shipped billing before we had users to bill, which in hindsight was the wrong order.

Three languages. French, Arabic, English. Arabic is right-to-left, which touches every layout. Doing this before launch instead of after roughly doubled every UI task from April on.

Onboarding. Seven steps, photo uploads, WhatsApp number validation, city search, delivery areas. It's the most-touched part of the codebase because it's where new creators drop off, and every drop-off we saw became a fix.

Content. Twenty-eight blog posts before launch, then two a day after. Not code, but it's on the calendar and it competes with code for the same hours.

What was fast

The visible product. Leaderboard, profiles, dashboard: seven weeks, and that includes the waitlist phase. With a design system and Next.js, screens are quick. It's the decisions around them that aren't.

Deploys. Vercel from week one. Every branch had a preview URL. Never spent a day on infrastructure.

Mobile. The Expo app started August 9 and had real screens the same day, because the API and the design already existed. Second platforms are cheap when the first one is done properly. How the app is put together is in Expo Router vs React Navigation.

So how long should your app take?

If you're a founder paying someone to build a first version, not nineteen weeks. MeltsWell took that long because it was ours, and because we built the second and third layers before anyone used the first one.

A first version for someone else, done the way we'd do it now:

What	Weeks
Landing page and waitlist	1
Small tool: sign-in, one core flow, simple dashboard	2 to 3
Full product, first version	5 to 7
Two-sided marketplace like MeltsWell	6 to 9
Mobile app with backend	7 to 10

Those are our real ranges, from how much it costs to build an app. The MeltsWell equivalent is the 6 to 9 week row: the May version, in English only, no billing, orders on WhatsApp. That's what we should have launched with. Billing and languages come after people show up.

Three rules from the log

1. Ship the May version. The first thing that works end to end, in one language, with no payments. That's what an MVP is. If nobody uses it, the other twelve weeks didn't matter.
2. Billing after the tenth user. You don't know what to charge until you've watched ten people use it for free. We wrote the pricing spec twice because we guessed the first time.
3. Count commits per month, not total. 472 commits in June looked like progress. It was mostly rework. The 150 in May were the product.

If you want to see what your version would cost and take, the cost calculator gives a range in a minute. Or tell us what you're building.