A marketplace is the most attractive business model to describe and the hardest to start, because on day one it is worth nothing to either side. Buyers arrive, find nothing, and leave. Sellers list, sell nothing, and stop. The engineering is largely a solved problem — catalogue, search, orders, payments, ratings — and building it well does nothing whatsoever to solve the actual problem, which is that an empty market has no value.
The first decision is which side to solve manually. Almost every successful marketplace began by faking one side: recruiting sellers individually by phone, seeding listings by hand, or serving the first hundred transactions through a spreadsheet and a WhatsApp group. This is unglamorous, does not scale, and is the correct thing to do, because it produces real transactions from which you learn what the product should be. Teams that build the full platform first and then look for participants have built a specification based on assumptions.
Which side to seed depends on which is scarcer and harder to replace. In most categories supply is the constraint and demand is buyable, so the early work is signing sellers and making them successful, one at a time, with the founder's phone number as the support channel. In categories where supply is abundant and attention is scarce, the reverse holds. Getting this backwards means spending your budget on the easy side and failing on the hard one.
Narrow the market until it can be dense. A marketplace serving one city, one category and one type of buyer can achieve enough density to be genuinely useful with a few hundred participants. The same number spread across the country and twelve categories is uselessly thin, and every visitor's experience is an empty search result. Liquidity in a small market beats presence in a large one, and expanding after density is achieved is far easier than achieving density after expanding.
Then watch the two numbers that describe whether the market works. What proportion of searches or requests result in a match, and how long a matched transaction takes to complete. Those tell you whether the marketplace is a marketplace or a directory. Gross volume flatters you at this stage because it can be bought with discounts; match rate cannot be, and it is the leading indicator of whether anyone will still be here when the incentives stop.
The final risk to design against is disintermediation — both sides meeting through you and then transacting privately. Preventing it by force rarely works, and the sustainable answer is to keep providing something they cannot get by going around you: payment protection, dispute resolution, logistics, discovery of the next transaction rather than this one. If the honest answer is that you add nothing after the introduction, the business is a lead generation service, and it is better to price it as one than to discover the fact through leakage.
