How to build a marketplace app: what shipping MeltsWell taught us

MeltsWell is a marketplace for home bakers and small sweet shops. Bakers get a page and a place on a leaderboard. Buyers find them and order on WhatsApp. We started it on March 25, 2026, launched in August, and had 59 creators across 18 cities by mid-September.

This is what building it taught us about building a marketplace app, in the order you'll hit the decisions.

1. Pick the side you build for first

A marketplace has two sides and you can't build for both at once. Pick the side that's harder to get and whose presence brings the other side on its own.

For us that was the bakers. A buyer with nobody to buy from leaves and doesn't come back. A baker with a good page has a link to share on Instagram today, even with zero buyers on the platform. So the baker gets value on day one, and every baker who shares their page brings buyers with them.

Test for your own marketplace: can one side get something useful before the other side exists? Build for that side first. For MeltsWell the "single-player" value was a nice page with a WhatsApp button. For a booking marketplace it might be a free scheduling tool. For B2B it might be a quote generator.

2. The four features a marketplace MVP needs

Everything else waits.

Feature	What it did on MeltsWell	When it shipped
Seller onboarding	Seven steps: business details, city, photos, products, how to reach you, go live	April
A public listing page	The baker's page with their best creations and a share link	Early May
Discovery	The leaderboard	May
A way to transact	A button that opens WhatsApp or Instagram to order	Before launch, reworked in August

Plus an admin review panel, because in a marketplace one bad listing makes everyone look bad. Every new MeltsWell page is reviewed before it goes live. At our size a person does it. That was the right call and it still is.

Things we didn't build at first, and you shouldn't either: reviews, in-app chat, delivery tracking, in-app payments, a mobile app, and seller analytics.

3. Don't handle the money until you have to

The default marketplace model is a take rate: payments go through you and you keep a cut. It's also the most expensive thing to build: checkout, refunds, disputes, payouts to sellers, tax, and in many countries a licence.

We didn't. Moroccan bakers already take orders on WhatsApp and get paid in cash or by transfer. Forcing that through our checkout would have made the product worse for both sides. So the "transaction" on MeltsWell is a button that opens a chat, and we make money from subscriptions and paid placement instead. The whole reasoning is in how free apps make money.

If your buyers and sellers already have a way to pay each other, let them. Hand off, measure the hand-off, and charge sellers for the value you add. Build payments when the hand-off is the thing losing you orders.

4. Discovery is the product

On the buyer side, a marketplace is a search and ranking problem. Two things we learned:

New sellers need a way in. A pure popularity ranking buries anyone who just joined. We added a "recently joined" rail and a launch cushion for new pages in July, so a baker's first week on the platform isn't their last.

Local means local. In September we made the leaderboard Morocco-only. Foreign creators keep a live page but don't rank, because a buyer in Casablanca can't order from Paris. If your marketplace is physical (food, services, rentals), filter by place before anything else.

The ranking score updates nightly from analytics events: views, order taps, product clicks. A daily score is plenty for a leaderboard and far simpler than a live one. More on that setup in PostHog vs Mixpanel.

5. The stack

For a marketplace in 2026, this held up well:

• Next.js on Vercel for the web app, public pages and API. Seller pages render on the server, so they rank in Google.
• Supabase for the Postgres database and image storage.
• Clerk for sign-in.
• Stripe for seller subscriptions and one-time featured spots.
• Resend for email. We send 13 transactional emails and nothing else (Resend vs SendGrid).
• PostHog for analytics, and the source of the sellers' own dashboards.
• Upstash for rate limiting on sign-in codes and other endpoints people like to abuse.
• Expo with Expo Router for the mobile app, started in August on top of the same API.

None of this is exotic, and that's deliberate. A marketplace has enough hard product problems without adding hard infrastructure problems.

6. What it costs to build a marketplace app

Our price for a two-sided marketplace MVP, the MeltsWell-sized version in one language without billing, is $15,000 to $25,000 over 6 to 9 weeks. Adding a mobile app on top brings it to the web + mobile range in the cost calculator.

MeltsWell itself took 19 weeks to public launch because we built billing, three languages and a lot of polish before anyone used the first version. The month-by-month breakdown is in how long it takes to build an app.

7. What we'd do differently

Launch the May version. Onboarding, pages, leaderboard, WhatsApp button, admin review. English only. That was a working marketplace, and it sat unlaunched for three months while we built billing and translations.

Onboard the first sellers by hand, sooner. The first six creators went live in July with a lot of help from us. Watching a real baker fill in the form is the fastest way to find what confuses them, and we could have done it in April.

Charge later. We wrote the pricing twice. The first version had no free tier and would have killed supply: a leaderboard with six bakers isn't a marketplace. Watch real sellers use it for free, then decide what's worth paying for.

Keep the seller page shareable from day one. Every baker who shares their MeltsWell link on Instagram is free marketing for both sides. It's the cheapest growth channel a marketplace has, and it only works if the page looks good enough to share.

If you're building one

Write down your two sides, the single-player value for the first side, and how a transaction happens without you touching the money. If you can fill in those three lines, you have a marketplace MVP scoped. The general version of this, for any product, is in MVP development for startups.

And if you want it built by people who've shipped one, tell us what you're building.