पाठशाला Pathshala · हिसाब Hisāb, Unit economics · Lesson 04 · Start

Marketplace unit economics: take rate, GMV and the leaky bucket

GMV is the biggest number in the building and the one the platform never owned. How to get from GMV to net revenue to contribution, and the subsidies, incentives and leakage that most decks leave out.

Pathshala, The Founder Library · 11 October 2026 · 8 min read

A marketplace deck opens with gross merchandise value because GMV is the largest number in the building. It is also the one number on the slide the company never owned. The buyer paid it, the seller received most of it, and the platform kept a slice. Marketplace unit economics is the discipline of following that slice from the transaction to the bank, and of counting everything the platform spent to make the transaction happen.

This lesson sets out the three numbers in order, GMV, net revenue and contribution; states what take rates have looked like at the companies that published them; names the six leaks that sit between net revenue and contribution; and gives a figure in which every one of them can be moved. It closes with the monthly ledger a marketplace founder should keep.

GMV, take rate, net revenue

GMV is the total value of transactions that passed through the platform in a period. Net revenue is the share the platform kept: commission, listing fees, advertising sold to sellers, delivery fees charged to buyers, subscription fees. The take rate is net revenue divided by GMV. Andreessen Horowitz’s 16 Startup Metrics puts the first rule in five words, GMV does not equal revenue, and defines revenue as the portion of GMV the marketplace takes. Anu Hariharan made the same point at Y Combinator’s Startup School in 2019 with Airbnb: a $200 booking is not $200 of net revenue, because Airbnb does not make the full $200.

Two refinements before the number is honest. First, compute take rate on completed GMV, not booked GMV. An order that was cancelled, refused at the door or returned to origin earned nothing and may have cost a delivery; a services booking that the buyer and seller settled in cash after meeting once through the app earned nothing at all. Second, decide what counts as revenue when the platform itself holds inventory. A quick-commerce company that buys the goods and sells them is a retailer for that part of its business and its revenue is the full sale price with cost of goods beneath it; mixing that with the commission model inflates the take rate without anyone intending to.

What a take rate can be

Bill Gurley’s 2013 essay A Rake Too Far is still the best single reference on where take rates sit and what happens when they are set wrong. His figures: eBay at roughly ten per cent, Amazon’s marketplace at six to fifteen per cent by category and about twelve on average, Apple and Facebook at thirty on digital transactions, Groupon near thirty-eight before the merchant’s own discount. His argument is that the most dangerous strategy for any platform is to price too high, because a high rake pushes up prices for buyers, drives suppliers to look elsewhere and creates the market for a competitor. His example of the opposite move was oDesk, which cut its commission from thirty per cent to ten, lost two-thirds of its near-term revenue and went on to overtake its nearest rival.

The rake also moves with power. The 2023 edition of the a16z Marketplace 100 noted that Amazon’s fees as a share of product price rose from 35 to 52 per cent over six years once advertising and fulfilment were counted. A seed marketplace should read both halves of that. In its first years it has no power and the rake must be low enough that supply wants to be there; the right to raise it is earned by becoming the place demand goes, and the companies that raised it before earning it are the ones Gurley’s essay is about.

The six leaks between net revenue and contribution

Net revenue is what the platform earned. Contribution is what it kept after every cost that existed because the order existed. Swiggy’s Q4 FY25 shareholder letter defines it precisely for food delivery, adjusted revenue less delivery and other charges less platform-funded discounts less other variable costs, and reports it as a share of gross order value. That definition is a list of the leaks. Here is the full one.

Platform-funded discounts. A coupon that the seller does not pay for is a cost of the platform, and it is the single largest item most seed decks leave out. Delivery subsidy. The customer pays ₹30 for delivery and the rider costs ₹60; the ₹30 gap is a cost of every order. Payments and refunds. The gateway charges on the gross value of the transaction while the platform earns on its slice, so a two per cent fee on GMV is eleven per cent of an eighteen per cent take. Support and disputes. Every order that generates a ticket costs a person’s time. Supply incentives. Rider joining bonuses, seller onboarding credits and guaranteed minimum earnings are acquisition costs for the other side of the market and belong in this list, not in marketing. Leakage. Cancellations, returns to origin and transactions that moved off the platform cost a delivery or a lead and earned nothing.

Each leak is small as a share of GMV, which is why they hide. Two per cent here and three per cent there against a GMV of ₹20 crore a month sounds like rounding. Against the ₹3.6 crore the platform actually keeps at an eighteen per cent take, five per cent of GMV is ₹1 crore and more than a quarter of net revenue is gone.

One month, worked

A home-services marketplace in Pune moves ₹2 crore of bookings a month and charges service providers eighteen per cent. Six per cent of bookings are cancelled or completed in cash after the first visit, so completed GMV is ₹1.88 crore and net revenue is ₹33.8 lakh. The platform funds first-booking discounts worth four per cent of GMV, ₹8 lakh. It pays its providers a travel allowance that the customer’s convenience fee does not cover, three per cent of GMV, ₹6 lakh. Gateway, refunds and a support desk run at two and a half per cent of GMV, ₹5 lakh. Contribution is ₹14.8 lakh a month, 7.4 per cent of GMV, and the take rate after discounts is 12.9 per cent rather than the eighteen on the slide.

The figure opens on the Pune company. Push the discount slider to twelve per cent, which is what a festive-season acquisition push looks like, and contribution crosses zero. Push leakage to fifteen per cent, which is what happens to a services marketplace when customers learn they can book the same plumber directly, and a fifth of the contribution goes with it while every cost stays. Then move the take rate up to twenty-five per cent to fix it and ask whether the plumbers would stay.

Reading the result

Three numbers come out of the figure and each has a use. Net revenue is what to build CAC and LTV on; the lesson on [CAC, LTV and payback](/library/cac-ltv-and-payback-the-three-numbers) explains why a marketplace that computes payback on GMV has computed nothing. Contribution as a share of GMV is the number listed Indian marketplaces now report and the one an investor will recompute from your data. Swiggy’s food delivery stood at 7.8 per cent of order value in Q4 FY25 after a decade; Instamart stood at minus 5.6 per cent; Blinkit reported 2.4 per cent in Q3 FY24 in its second positive quarter. A seed marketplace at two per cent of GMV needs ₹50 crore of monthly GMV to earn ₹1 crore of monthly contribution, and should know how many years that is. Take rate after discounts is the honest price the platform is charging, and if it is half the headline the headline is a hope.

GMV is what moved through the platform. Net revenue is what stopped there. Contribution is what was left after you paid to make it move.

Why Indian decks still lead with GMV

Between 2013 and 2015 the largest e-commerce companies in India ran on GMV targets and said so, and the question of profitability was treated as a distraction from market share. The sector spent the years after 2016 unlearning that, and the listed survivors now publish contribution by segment every quarter. The habit survives at seed stage because GMV grows fastest when a platform subsidises both sides, and a slide that shows GMV tripling while contribution falls can be built in a way that shows only the first line.

Do not build that slide. The investor will recompute contribution from the order data in diligence, and the founder who did it first, showed the negative number and showed the dated route to a positive one is the founder who gets the term sheet. The one who showed GMV and hoped is the one who gets the question in the room.

The monthly leak ledger

On the fifth working day of each month, once the previous month’s orders are settled, build one table. Rows are the stages: booked GMV, completed GMV, net revenue, each leak in turn, contribution. Columns are the months. Add a second copy of the table per city or per category, because a platform that is contribution-positive in Bengaluru and negative in three new cities has a different problem from one that is negative everywhere. Then compute two per-order numbers: the platform’s take per completed order and its total incentive per completed order, discounts and supply payments together. When the second is larger than the first the platform is paying people to transact, and the month in which that reverses is the month the business begins.

Decide one thing from the table each month. The ledger is twenty minutes once the data is clean. Cleaning the data, which means tagging every rupee of discount and incentive to an order, is the real work and is also the thing an acquirer or a Series A lead will ask for first.

Nothing here is legal, tax or investment advice. It is a way of counting, and the companies cited publish the count every quarter so that you can check yours against theirs.

Sources

1. Bill Gurley, A Rake Too Far: Optimal Platform Pricing Strategy, Above the Crowd, April 2013 — Take rates at eBay, Amazon, Apple, Facebook, Groupon and oDesk; the danger of pricing too high.
2. Jeff Jordan, Anu Hariharan, Frank Chen and Preethi Kasireddy, 16 Startup Metrics, Andreessen Horowitz, August 2015 — GMV does not equal revenue; revenue is the portion of GMV the marketplace takes.
3. Y Combinator, Startup School Week 3 Recap: Anu Hariharan and Adora Cheung, September 2019 — The Airbnb example; net revenue is the cash that reaches the bank.
4. Swiggy Limited, Q4 FY2025 Shareholder Letter, May 2025 — Definition of contribution margin; food delivery at 7.8 per cent of GOV and Instamart at minus 5.6 per cent; Instamart GOV growth of 101 per cent.
5. Inc42, Zomato Q3: Blinkit Posts Second Consecutive Contribution Positive Quarter, February 2024 — Blinkit contribution margin of 2.4 per cent of GOV in Q3 FY24.
6. Olivia Moore and Zach Cohen, The a16z Marketplace 100: 2023, Andreessen Horowitz — Amazon’s fees as a share of product price rose from 35 to 52 per cent over six years.
7. Itika Sharma Punit, Indian e-commerce dumps its once-favourite baby GMV, Quartz India, February 2017