Building marketplace platforms that survive scale

I have worked on marketplaces at four very different scales: eBay's seller platform, Google Shopping Express, Indigo Ag's grain marketplace, and Wish. They failed and succeeded in different ways, but the tensions were the same every time, and they were always resolved in the same order.

Liquidity is the only thing that is genuinely urgent

Early on, nothing else matters. A marketplace without liquidity is a directory, and no amount of product quality rescues a directory. This justifies a great deal of unscalable behavior — manual matching, subsidised transactions, concierge onboarding — and teams are usually right to do it.

The mistake is not doing the unscalable thing. It is failing to write down when you will stop. Liquidity tactics have a way of becoming permanent cost structures because no one wants to be the person who turned the growth engine off.

Trust is slow to build and instant to lose

Trust is the asset that compounds and the one most easily spent for short-term liquidity. Every marketplace faces the temptation: loosen seller standards, relax listing quality, let the take rate slide, and watch the numbers improve for two quarters.

At Wish, rebuilding trust was the entire job. Refund rate at 40% is not a support problem, it is a statement about whether the marketplace works at all, and the same is true of a 35-day time to door. Getting those to 3.5% and 10 days respectively is what moved NPS from negative to +40 — and it also increased order size by 33%, because people who trust a marketplace buy more from it. The trust work and the revenue work were the same work.

Refund rate at 40% is not a support problem. It is a statement about whether the marketplace works at all.

Unit economics are the constraint that ends the argument

Liquidity and trust are both purchasable, at a price. Unit economics determine how long you can keep paying it, which makes them the constraint that eventually resolves every strategic disagreement — usually later and more painfully than if they had been treated as a design input.

On Google Shopping Express, the pivot that mattered was to the merchant experience rather than to the consumer product: a path with much higher scalability, a step change in the ability to onboard and manage merchants, and a breakthrough in product data quality. That work reduced transaction cost by more than 90%, and the 90% is what made the partnerships with Target, Costco, Walmart and Albertsons viable. The consumer feature set was downstream of the economics, not the other way round.

Pick which side of the market you are actually serving

Marketplaces claim to serve both sides. In practice, every consequential decision favors one, and the ones that survive are explicit about which.

At Indigo Ag, the pivot was precisely this. The company had been operating as a broker — effectively competing with the grain buyers it needed as partners. Moving to an open marketplace between grain buyers and farmers, and firmly acting as a SaaS provider rather than a counterparty, reestablished the partnership and unlocked the first enterprise client adoption. The technology mattered, but the decision was about which side of the market the company was on. The full technology and organizational pivot took two months; the clarity that made it possible was the hard part.

The platform decision underneath all of it

At eBay, the durable structural choice was one central inventory system behind everything — cross-border visibility, Buy Online Pickup in Store, multi-channel selling, all powered by the same source of truth through feeds, APIs and integrated experiences. It ended up powering 100% of the marketplace's transactions, and it is what made onboarding merchants like Target and Newegg possible at all.

That is the shape of a platform decision that survives scale. Not a feature, and not an architecture diagram — a single source of truth that every surface is obliged to go through, chosen early enough that the obligation is cheap to honour.

What I would tell a team starting now

• Buy liquidity deliberately, with a written stopping condition. Undated subsidies become permanent.
• Treat trust metrics — refund rate, delivery time, listing quality — as revenue metrics, because they are.
• Make unit economics a design input rather than a quarterly surprise.
• Decide which side of the market you serve, and be honest when a decision favors the other one.
• Choose the single source of truth early. Every quarter you wait multiplies the cost of imposing it.

None of these are novel. They are simply the tensions that show up in the same order at every scale, and the teams that do well are the ones that recognize which one they are actually arguing about.