The Two-Sided Marketplace Cold Start: Which Side Do You Build First

Why Two-Sided Markets Are Genuinely Hard

Two-sided markets are hard not because the product lacks features. They are hard because what you sell is the other side. Suppliers arrive when they see demand, buyers arrive when they see supply. On day one, with both sides empty, even a well built interface shows nothing but a blank list. That condition is the cold start.

Single-player software delivers value from the first day. A bookkeeping tool organizes your books whether or not anyone else uses it. Matching services work differently. When a new user signs up and sees no results, they do not conclude that the product is broken. They conclude that they came to the wrong place. Then they leave and do not return.

This is the point founders miss most often. Bringing back a user who saw an empty list costs more than acquiring that person in the first place. That is also why spending on ads to push both sides in simultaneously is risky. The budget goes out twice over, and if one side arrives even slightly late, the side that came first has already gone. Worse, the people who left tell others that the service had nothing in it. First impressions happen once, and in a two-sided market that impression is usually decided by listing density.

The second trap is the aggregate number. There is a long stretch where signups grow while completed transactions do not. If you have a hundred suppliers in Hanoi and a hundred buyers in Ho Chi Minh City, the dashboard reads two hundred and nothing closes. The real metric in a two-sided market is not total accounts but the number of results that survive the filters a real person actually applies. Once someone narrows by area, price band, timing and language, how many rows remain on screen. While that number stays in single digits, nothing you bolt on turns into growth.

The third trap is the time axis. The two sides have unequal patience. The side that came to earn money will wait a few weeks. The side that came to solve a problem right now will not wait a few minutes. If you refuse to acknowledge that asymmetry and decide to gather both sides at once, you build a machine that keeps acquiring the impatient side and keeps losing it. The budget drains and no learning accumulates.

Cold start is therefore a sequencing problem rather than a marketing problem. The decision is which side to fill first, and what the other side will not be shown in the meantime. Postpone that decision and the product keeps getting built while the market never opens. Most services that ship complete and then see nothing happen have failed at sequencing, not at features.

How to Decide Which Side to Gather First

Replace intuition with four questions. Which side needs you more, which side is smaller and therefore easier to find, which side will wait without results, and which side leaves something visible for the other simply by registering. When all four answers point the same way, fill that side first.

The first question is intensity of motive. The side that is short on something moves first. A small business with empty slots has a clear reason to try a new channel. A shop already fully booked will not spend time on an unfamiliar platform. The suppliers you need early are not the strongest operators in the market but the ones who want to fill the hours they currently have open. Miss that distinction and your sales difficulty multiplies.

The second is recruiting difficulty. One side is already enumerated and the other is scattered. Shops have signage and addresses, academies and tutors are already registered somewhere, hiring companies are already posting. The individual consumers on the other side are much harder to locate as a group. The side with an existing list is cheaper to reach. Having a list, though, is not the same as being easy to persuade. A list only makes contact cheap, and persuasion depends on the next criterion.

The third is patience. Ask which side can stay while the screen is still empty. The side with an expectation of income generally stays. They register, forget about it, and respond when contacted. The side with an immediate need does not stay. There is no way to tell someone who needs an interpreter tonight to wait until next month. So accumulate the side that can wait, and call the side that cannot only after the accumulation is done.

The fourth is the inventory effect. One side becomes visible the moment it registers and the other does not. When a supplier creates a profile, that profile immediately becomes a listing, a search result, and an indexable page. When a buyer registers, the record sits in the database and leaves no trace on screen. You have to gather the side that fills the screen first, so that the next visitor has something to look at.

Some markets return conflicting answers. Where the buying side is small and identifiable, as with specialized equipment rental or large subcontracting structures, demand is the scarce and enumerable side. There it is faster to lock in the buyers first and pull supply toward their requirements. The principle is not supply first as a rule. It is that the scarce, patient, screen filling side goes first.

One more point belongs here. Early on, a human can stand in for an entire side. When a request arrives, an operator reaches out by hand, finds a counterpart and connects the two. Making a few matches manually before automating anything teaches you which conditions actually govern a match. This is usually called the concierge approach. The conditions you learn at that stage become your filters and sort orders later. Skip the stage and you will build a wall of filters nobody uses.

When Supply First Is Usually the Right Call

In most matching services, suppliers come first. They are fewer and easier to find, they tolerate an empty screen because they expect income, and the act of registering produces listings and search results. When those three conditions overlap, supply first is the answer. There are clear exceptions, and they are worth checking before you commit.

The classic case for supply first is location based service. In markets where physical place and people are bound together, such as salons, workshops, restaurants, tutoring, construction and translation, the supplier list is itself the content. As listings accumulate, an inbound path from search appears, and that inbound traffic then lowers the cost of recruiting buyers. Gather buyers first and this loop has no starting point, because you are buying attention before you have anything to show.

The common mistake when gathering suppliers is chasing count alone. Profiles collected for the sake of registration lengthen the list without producing transactions. Once the list contains suppliers who cannot be reached, profiles with no stated conditions, and pages carrying information from several months ago, buyers fail twice and leave. In the early phase, response rate matters more than registration count. Twenty live profiles beat a hundred dead ones. It is often better to secure a commitment to respond before inviting an early supplier to register at all.

The exceptions are equally clear. First, markets where suppliers are already listed on several platforms at once and switching costs are effectively zero. There, accumulating supply produces no differentiation, because the same listings exist everywhere and the contest shifts to whoever holds demand. Second, digital goods markets where supply is effectively unlimited and replication costs nothing, so scarcity of supply is not the bottleneck. Third, the case described above, where demand is the scarce and enumerable side.

Vietnam adds a layer to the judgment. A large share of suppliers already sell through Facebook pages or Zalo groups, and those channels still work for them. Registering on a new platform reads as extra work. In that situation, promising new customers is not enough to gather suppliers. A promise carries no value until it is verified, and verification takes time. What you hand them during that interval is the subject of the next section.

The same judgment repeats in the services Yeowubie Interaction builds. Job Connect VN connects hiring companies with job seekers, and Langtori is a matching and mapping service between people learning Korean and people teaching it. In both cases the side to fill first was the side that already existed on a list and held an expectation of income. The teaching itself is delivered by the individual tutors and institutions participating on the platform, while what we build is the tool that lets the two sides find each other. That boundary is preserved in the product design as well.

Giving One Side a Useful Tool Before Opening the Market

The approach is to ship something one side finds useful while the other side does not yet exist. If today's work gets lighter without any counterpart present, people stay. And when the information accumulated in that tool later converts into listings, you open the market from a screen that is already populated.

A widely known example is OpenTable. Most people know it as a consumer reservation network, but it began by selling reservation management systems to restaurants. A restaurant needs to manage its own booking book whether or not diners arrive, and the consumer side only became meaningful once enough restaurants were running that tool. Reversed, diners would have opened a screen with almost no restaurants available to book.

Three things have to hold for this to work. First, the tool must genuinely reduce work today without the other side, because an expectation of future customers will not keep anyone using it. Second, data has to accumulate as a natural byproduct of use, because nobody will complete a separate data entry chore. Third, that data must be in a form that converts into listings when you open the market. If photos, price conditions, available hours and location are already organized, flipping the switch produces a searchable list immediately.

The structure of Langtori illustrates the pattern. According to its landing page, the service matches by region, level and purpose, lets tutors organize class schedules and track progress, and lets each person share their own class page on social media. The scheduling and the class page are the parts that create value for the teaching side even when no learner has been connected yet. Organize your introduction and conditions once and inquiries get shorter to answer, and that organized information becomes exactly what a learner sees in search results. To be clear again, the teaching is provided by the individual tutors and institutions who participate, and what we build is the tool that helps them find and organize.

There is a simpler example of a tool with standalone value. onSpots records and manages attendance and works completely within a single organization, with no network effect required. What a team preparing a matching service should take from it is not the feature set but the test. Ask what value remains in your product if you assume the other side never shows up. If nothing remains, then advertising is the only way you have to survive the cold start.

The risk of a tool first approach deserves a mention. If the tool succeeds too well, opening the market gets postponed indefinitely. There are cases that run for years, where the tool brings in modest revenue and users are satisfied while the matching side never opens. The guard against this is to fix the transition condition numerically at the outset. Decide that once a given area reaches a set number of active suppliers you open the demand side, so that the condition decides rather than a person.

Narrowing Geography and Category to Reach Critical Density

Two-sided markets run on density, not total size. A transaction happens when someone enters real conditions and gets back several usable results. So in the early phase you narrow the market rather than widen it. Build density first in one city, one district, one category.

There are usually three axes to narrow along: geography, category and situation. Not all of Hanoi but a few districts, not all languages but beginner conversational Korean, not all hours but weekday evenings. Narrowing all three at once makes the market look small enough to be unsettling. Yet the experience of a search result that comes back full inside that small zone is your first real asset. When transactions repeat inside a narrowed zone, you have a pattern to replicate. Transactions scattered thinly across a wide market leave no pattern behind.

Several widely reported cases illustrate the method. Uber did not launch nationwide at once but opened city by city, refilling supply in each city. Airbnb concentrated early on particular cities and particular periods to gather listings. Neither tried to populate a national map from the start. What can be cited here is the publicly described approach, since the internal figures of either company are not something we can know.

Executing the narrowing calls for an experience standard rather than a target number. Instead of aiming to gather fifty suppliers, set the standard that five representative search conditions each return at least five rows. The first goal can be met with the screen still empty, while meeting the second guarantees a full screen. Set that standard and it also settles which suppliers to recruit next, because every search condition that comes back thin is your next sales list.

Define the signals for expansion in advance too. Watch whether transactions continue after the operator stops connecting people by hand, whether a buyer who transacted once comes back, and whether suppliers update their own pages without being asked. Once those three start sustaining themselves in one zone, replicate the same procedure in the next. If all three still depend on the operator's hands, adding a second zone simply spreads those hands too thin and both zones fall over.

Finally, a word on the psychological difficulty of narrowing. In front of investors and inside the team alike, a small market looks like a small ambition. So everyone wants to talk about the whole country and every category from the beginning. But starting wide in a two-sided market means choosing a state where every search condition returns too few results. Starting narrow is not a decision to give up the large market. It is a decision about which zone comes first on the way there. It orders the sequence, it does not shrink the ambition.