This book was recommended by Vladimir Pravdivy, who spent many years as the CEO of Avito (Avito is Russia's largest classifieds platform, roughly the local Craigslist) and grew the company into the giant we know today. As so often happens, the book sat waiting for its moment for two years — but the wait was worth it :) It's a slim little thing (70 pages), more of a handbook than a book, really, but it explains with remarkable concision how to think about marketplaces.

The book is freely available online — you can download it as a PDF or EPUB. Below is a brief summary, along with my thoughts.

What is a marketplace? Picture a market where deals (transactions) are struck between two parties. For example, therapists are looking for clients, and clients are looking for therapists. Or passengers are hunting for cabs, and cabs are hunting for passengers. This market already has some transaction volume happening without you. Now you show up with your platform and say to the parties in the deal: let me help you out. They start using you, and from then on some share of the deals goes through with your involvement.

And there you are, sitting astride the pipe through which the money from all these closed deals flows. And you want to know two things: 1) how much money is flowing through the pipe, and 2) what share of that flow goes to your platform.

The first is called Gross Merchandise Value (GMV), the second — take rate. The book argues that these two metrics are the key ones for any marketplace.

The whole spectrum of existing marketplaces lays out beautifully along these two measures — and marketplaces can be wildly different from one another.

Take ride-hailing services: Yandex.Taxi, Uber, Lyft. The market has two sides — passengers and drivers. The service takes care of everything: finding a car for the passenger and a passenger for the driver, processing the payment, plotting the route, making sure there are enough cars in the city, checking the condition of the cars and the competence of the drivers. The services embed themselves so deeply in the transaction that it's fair to ask: aren't the drivers effectively employees by now? Thanks to this degree of control, the services can command a take rate of 20–30% and earn enormous revenue on a comparatively modest GMV.

Now for an example from the opposite end of the spectrum: classifieds boards, such as Avito, Craigslist, Thumbtack. There is often no clear line between buyer and seller at all: today you show up as a buyer, tomorrow you decide to clean out the garage and become a seller. The platform barely controls the transaction; it merely brings buyer and seller together, and how they get in touch, where they meet, and whether the deal actually happens is their own business. But the platform is enormous: you can find everything from diapers to Swiss watches, from cashier job listings to entire businesses for sale. The result is a huge GMV (a couple of years ago Avito boasted a GMV equal to 3% of Russia's GDP) but a rather modest take rate, earned not on transactions but on listings.

Exercise: think about where italki, Booking, BlaBlaCar — and any other intermediary services you can think of — would land on these two axes.

What does all this mean for data analytics? Unit economics is no longer so easy to compute, because there is no sale: we aren't the ones doing the selling, we aren't the ones providing the product or service — we merely connect the two sides (though, as we've seen, the depth of that "merely" varies). Accordingly, our commission, or the price of our additional services, depends on the state of the marketplace: how many sellers there are (supply), how many buyers (demand), and whether both sides have enough liquidity.

So we can still measure the funnel, acquisition cost, and retention metrics for each side of the marketplace, but to estimate LTV we need to know the state of the other side of the market over the user's entire lifetime in the service. That complicates things considerably.

That's why marketplaces try to devise some quantity that captures each user's contribution to the market on the platform while being decoupled from the other side. For taxis, this could be the number of hours a driver is available online (regardless of whether there was a ride during that hour); for a classifieds board — users weighted by their activity on the board. And then unit economics gets calculated against that quantity.