No. 02 Foundations
What a network effect does
In most businesses each new customer costs more than the last. A working network effect reverses that, across acquisition, retention, pricing and margins.
The last essay was mostly about what a network effect is not. It makes readers cautious. But the caution has a reason, and the reason is how valuable the real thing is. A working network effect does something very little else in business does. It makes growth cheaper the more of it you have.
Most businesses run the other way. The first customers are the ones who wanted the product most and found it on their own. Every customer after that is a little harder to reach. Ad auctions get more expensive as you bid deeper into audiences who care less. A new city needs a new sales team. A new segment needs features, the product doesn’t provide yet. The cost of the next customer rises with size, which is why most growth curves bend down eventually, however good the product.
A network effect bends the curve the other way. Each new user makes the product better for the users already there, which makes the next user easier to win, easier to keep and easier to charge. That is what it means for a network effect to compound. It shows up across income statements, and in times on the balance sheets too.
Customers start arriving on their own
When a network is working, the product itself becomes the main reason new users show up. A traveler books on Airbnb because the listings are there, and a host lists because the travelers are there.
You can see this in Airbnb’s filings. In 2019, a normal year, about 23% of its traffic came from paid performance marketing, so roughly three quarters came directly or through unpaid channels. In the first nine months of 2020 it switched off most performance marketing, and 91% of traffic still came in on its own.1 That year was distorted by the pandemic, and the company later went back to spending. But the point holds across the period: sales and marketing was 34% of Airbnb’s revenue in 2019 and 21% in 2025, while revenue grew about two and a half times.
IndiaMART, the business-to-business marketplace, made the same point. Its 2022-23 annual report said its 1,021 million visits that year were “entirely organic” and needed no significant marketing spend.2 Every supplier who lists adds products that buyers search for, and every buyer enquiry gives the next supplier a reason to list. (IndiaMART has since raised its advertising, which is a useful reminder that organic growth slows when a network nears the edges of its market.)
The difference with paid growth is what happens when spending stops. A business purchasing its growth has to keep purchase it. A network which has grown itself will keep the most even without spending.
Old customers get more valuable
You have probably noticed it yourself: you sign up for a new product, use it a lot for a week or two, and then forget about it. Most of its other users do the same. That is why, in most businesses, a cohort of customers is worth the most the day they signs up. After that their value decreases, as people use the product less and then eventually stop. Investors read a product’s retention curve before almost anything else.
In a working network the people under the influence circle of each user keep arriving after that user joined. The recruiter who signed up last year now has more candidates to search. The seller who listed in January has more buyers by December. The product they subscribe to keeps getting better without addition of new feature, so a cohort can actually be worth more in its second year than in its first.
Slack showed this in its 2019 listing documents. Its net dollar retention was 143%, meaning that a year on, the same group of paying customers spent 43% more. Slack usually arrives with one team. Every other team that joins gives the access to more colleagues, so the product gets more useful, and the organization pay for more seat. Later, same effect is stretched across companies with shared channels, which let an org work with its vendors and clients seamlessly.
The same loop runs in reverse, which is why networks can unravel as quickly as they compound. And there are ways to tell a real smile from a flattering one, which a later essay covers.
You can charge, and keep charging
Users pay for value they can’t easily get cheaper elsewhere. A network effect is a machine for producing exactly that kind of value, because the thing a user values is the other users, and a rival can’t copy those with a better feature.
Visa is the nice case. A cardholder carries the card merchants accept, and a merchant accepts the card cardholders carry. Neither side can leave alone. In its fiscal year to September 2026, Visa kept about two thirds of its $45.9 billion in net revenue as operating profit, before litigation provisions, and its margin is in the mid sixties a decade earlier too3. Mastercard, the same kind of network, ran an adjusted operating margin of about 59% in 2025. Margins like that, held for a decade, are what pricing power looks like when it is earned on both sides.
Smaller networks show the same thing in a quieter way. About 220,000 suppliers paid IndiaMART a subscription at the end of March 2026. They pay because that is where the buyers are.
The important qualifier is that the power sits with the side that gets the most value, and it holds only while the value is being delivered. Charging the wrong side, or charging at the wrong step, and you can suck the network. Who pays, and who gets subsidized is about getting monetization right.
Revenue that outgrows its costs
This is the consequence of first three. If customers arrive cheaply, stay longer and pay more, revenue grows faster than the cost of serving it. The product is the network, which grows through its users.
Airbnb again: between 2019 and 2025 revenue grew about two and a half times, sales and marketing grew about 1.6 times, and its adjusted EBITDA went from a loss of about 5% of revenue to a profit of about 35%.1 Craigslist, by AIM Group’s estimate, made about a billion dollars in 2018 with roughly fifty employees.4
Across many companies the pattern is consistent. Michael Cusumano, Annabelle Gawer and David Yoffie compared 43 platform companies in the Forbes Global 2000 with non-platform companies in the same markets. With similar revenue, the platforms had about half the employees and about twice the operating profit, and much higher market values.5 The venture firm NFX, looking at technology companies founded since 1994 that reached a billion dollars in value, estimated that about 70% of the value they created came from network effects.6 That is the firm’s own analysis rather than a peer-reviewed study, but it matches what the margins say.
Small differences matter here more than they look. On a business with a 10% margin, another 100 basis points is 10% more profit on the same revenue. A network effect tends to add those points year after year, because each year’s growth costs a little less than the last. Investors typically pay a higher multiple for earnings that rise this way, because they expect them to keep rising.
One network, sold four ways
LinkedIn shows what this looks like when a single network serves several markets at once. A professional keeps a profile current because recruiters look there. Recruiters pay because the profiles are current and professional is highly active so response rate is high. Consultants, startups and B2B sales teams go there for the same reason, to find the buyer at a company or to be found by one. And the feed gives members a reason to come back between job searches, which matters more than it sounds, because looking for a job is something most people do every few years, not every week.
Each new member therefore adds value to several groups at once: a candidate for recruiters, a prospect for salespeople, an audience for advertisers, a contact for everyone else. Microsoft describes four paid businesses built on that one network: Talent Solutions, Marketing Solutions, Premium Subscriptions and Sales Solutions.7 LinkedIn reported about $3 billion in revenue for 2015, the last full year before Microsoft agreed to buy it for about $26 billion. In Microsoft’s fiscal year to June 2026 that revenue grew another 11%, to roughly $20 billion, and the network had passed 1.4 billion members by Microsoft’s count.
That is compounding at another scale. Most networks compound inside one market. A network that several markets depend on compounds across all of them, and each new product it launches starts with an audience that is already there.
Compounding has to be set up
None of this happens because a business has users who could, in principle, be useful to each other. The loop only runs when several things are true at once, and most of the work of running a networked business is making them true.
- The segment has to be dense enough to hold. A network is worth only as much as the slice of it one user can use, and below a certain depth users leave before the next ones onboard. That is Density beats size.
- Users have to actually find each other. On a platform where both sides must agree, a recommender that ignores the other side’s interest wastes most of the network. See When both sides have to say yes.
- Ranking has to spread the value instead of concentrating it on the few users, otherwise users churn. That is Your ranking algorithm is your economic policy.
- Participants have to trust the other participants they meet, or every bad interaction teaches thousands to stay away (Trust is a network effect).
- The right side has to pay, at the right step (Who pays, and who gets subsidized).
- Acquisition has to buy users who add value for others, not just users who pay quickly (What ROAS can’t see).
Get these right and the four effects above arrive together, and keep arriving. Get one wrong and growth turns into expense.
Four lines on one chart
If you want to know whether your network is compounding, put four lines from the last three years on a single chart. The first is the share of new users who came without paid marketing. The second is the twelve-month retention of each year’s cohort, measured in revenue. The third is the average price you realized per paying customer. The fourth is your operating cost per active user.
If the network effect is working, the first three rise and the fourth falls, together. If only one of them moves, you are probably looking at something else, such as a brand, a sales team or a good year, and the previous essay will help you tell which.
Notes
Airbnb, Inc., Form S-1, filed 16 November 2020, sec.gov, for the traffic shares and 2019 figures; the S-1 notes that the company had suspended substantially all performance marketing in 2020. Airbnb, Inc., Form 10-K for 2025, filed 12 February 2026, sec.gov, for 2025. Adjusted EBITDA is the company’s own non-GAAP measure. Sales and marketing was 19% of revenue in 2024, so the most recent year moved up. ↑ ↑
IndiaMART InterMESH Ltd, Annual Report 2022-23, for the traffic statement, and Integrated Annual Report 2025-26, for paying suppliers. Its advertising expense rose from about 0.4% of revenue in 2024-25 to about 2.2% in 2025-26. ↑
Visa Inc., Form 10-K for the fiscal year ended 30 September 2026, filed 5 November 2026, sec.gov. Visa doesn’t report an operating margin; the figures here are calculated from its reported revenue and its non-GAAP operating expenses, which exclude litigation provisions. On a GAAP basis the margin was about 61%, and about 65% in fiscal 2015. Mastercard, fourth-quarter 2025 earnings release, 29 January 2026. ↑
Sara Salinas, “Craigslist posts annual revenue of $1 billion: study”, CNBC, 24 January 2019, reporting AIM Group estimates. Craigslist publishes no figures. AIM Group later estimated 2021 revenue at about $660 million. ↑
Michael A. Cusumano, Annabelle Gawer and David B. Yoffie, The Business of Platforms (HarperBusiness, 2019). The comparison uses medians for each group. ↑
James Currier, “70 Percent of Value in Tech is Driven by Network Effects”, NFX, first published in 2017, covering 336 companies founded since 1994 that reached $1 billion in value. ↑
Microsoft Corporation, Form 10-K for the fiscal year ended 30 June 2026, filed 29 July 2026, sec.gov, for the four lines of business and the 11% growth. The roughly $20 billion revenue figure (about $19.8 billion) and the member count, which Satya Nadella gave on Microsoft’s October 2026 earnings call, are from press coverage; Microsoft stopped reporting LinkedIn revenue separately from the following quarter. LinkedIn’s 2015 revenue was $2.99 billion; Microsoft announced the acquisition in June 2016 at about $26.2 billion. ↑