Build the harder side first — usually supply. Seed it manually, by hand, until it looks real, then open the easier side to it. The side that's expensive to fake (inventory, listings, providers) makes the product worth opening at all; the cheap-to-fake side (buyers, browsers, applicants) you can simulate yourself for the first few weeks.
Every two-sided marketplace MVP hits the same wall in week one: you need suppliers to attract buyers, and buyers to attract suppliers. Founders who guess wrong here don't fail loudly. They launch a marketplace with five listings and zero applicants, watch it read as abandoned, and burn the one shot they had at first impressions.
Which side should you build first?
Build whichever side is harder to fake convincingly. In most consumer and B2B marketplaces, that's supply — because supply requires real inventory, real listings, or real providers willing to show up before there's any proof the demand exists. Demand, by contrast, can be faked. You can be the first ten "buyers" yourself: message the suppliers directly, run the matching by hand, close the loop over email or WhatsApp instead of in-app.
Once supply exists and looks credible — twenty listings, not two — demand is usually two to ten times easier to bring on, because there's finally something worth showing up for.
Why does the harder side go first?
The logic is about what the marketplace makes credible, not what's technically simpler to build. A demand-side landing page with a waitlist is trivial to build and tells you nothing — anyone will join a waitlist for a product that doesn't exist yet. A supply-side listing is expensive to fake, because it requires a real provider to commit real time or inventory. When you've solved the expensive side, the cheap side follows.
This is also why "build both sides at 50% effort" is the failure mode. Half a marketplace reads as a dead one. A buyer who lands on a page with three sparse listings doesn't think "early stage" — they think "nobody uses this" and leave. Better to fully staff one side by hand and delay the other than to half-build both and let the platform look empty on day one.
What does a concierge MVP look like?
A concierge MVP is you, manually, standing in for the software. Before building a matching algorithm, a review system, or in-app messaging, run the match yourself:
- Recruit ten to twenty suppliers by hand. Cold outreach, not an application form. You're vetting quality, not scale.
- Bring demand in manually too, at first. Message people who fit the buyer profile directly. Route the first transactions through email, a spreadsheet, or WhatsApp — not a built product.
- Only automate the step that breaks under manual load. If you're doing five matches a day by hand, that's not a bottleneck yet. Automate the part that actually can't scale by hand, not the part that feels unfinished.
The product you eventually build should encode what you learned running the marketplace manually — not guess at it up front.
What have we seen in our own builds?
On HospiHealth, the original brief wanted a job-board marketplace: public listings, candidate profiles, a matching interface. We cut it in week one. A new hospital consulting firm with one employer and no candidate base yet would have launched a marketplace with five listings and zero applicants — worse than shipping no marketplace at all. We built a direct intake form instead; it got 43 real applications, and the marketplace stayed a roadmap item until there was volume to justify it.
Tirth, a pilgrimage travel platform, is the supply-first case worked all the way through. The pilgrim isn't looking for a booking platform — they're looking for a guide they trust. We built the guide side (supply) first: every itinerary authored by a named, vetted local guide, with testimonials and a phone number, before a single transaction flow existed. The guide network grew 4x by referral in the first quarter, because supply, once credible, recruits more supply.
How many suppliers do you need before you open demand?
There's no universal number, but the working rule is: enough that a new visitor doesn't think "empty." For most local or niche marketplaces, that's fifteen to thirty real listings with enough variety (category, location, price point) that a visitor's first three searches return something. Fewer than that, and you're better off staying in concierge mode — matching buyers to suppliers by hand — until the count clears the bar.
What's the rule that decides it?
Fake the side that's cheap to fake. Build, by hand if necessary, the side that's expensive to fake. Once that side is credible, open the other one — it will come faster than the first side did, and the product you build next will be built on a marketplace you've already watched work, not one you're still guessing about.
Heuristics
- Seed the expensive side manually before writing a matching algorithm. The algorithm is worth building once you know what a real match looks like.
- Half of both sides is worse than all of one side. A sparse marketplace reads as dead; a concierge-run single side reads as early but alive.
- Automate the step that breaks under manual load, not the step that feels unfinished. Most marketplaces can run five to ten matches a day by hand for weeks before automation earns its keep.
For the broader discipline of deciding what not to build in week one, see 4 builds, 4 cuts: what we didn't ship, and why each cut held.
Written 2026-09-03 by Naman Barkiya.