Part I Foundations

What a network effect is, what it does for a business, where it lives, and how it dies.

  1. Most network effects aren’t

    A network effect means each new user makes the product better for the users already there. Most products have something else instead.

  2. 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.

  3. Density beats size

    A network is only as big as the slice of it a single user can use, which is why a small player can beat a giant one city or community at a time.

  4. Networks unravel, too

    The loop that compounds growth also compounds decline. It starts in the connections between users, and it shows in replies long before it shows in MAU.

Part II Marketplaces & Algorithms

Matching, ranking, trust, pricing and acquisition: how good matches turn into retention, and retention into revenue.

  1. When both sides have to say yes

    On a shopping site a click is a vote. On a jobs or matrimony site it’s a request someone has to answer, and most recommenders are built as if it weren’t.

  2. Your ranking algorithm is your economic policy

    Ranking decides who gets seen, being seen decides who earns, and earning decides who stays. Run that loop for a few years and it picks your supply side.

  3. Trust is a network effect

    A scammer on a marketplace cheats a few people and teaches thousands to distrust strangers, which is why trust belongs on the growth dashboard.

  4. Who pays, and who gets subsidized

    In a two-sided market, who pays matters more than how much. How to pick the subsidy side and the money side, and the one place a paywall should never go.

  5. What ROAS can’t see

    Ad platforms find more of whoever you tell them is valuable. Send revenue from the first week and they find people who pay once, not the users who make the network work for others.

Part III Machine Learning

Data loops, graphs, and what AI does and doesn’t change.

  1. Your data moat is probably a learning curve

    More users, more data, a better product. That loop compounds only in a few places: the long tail, and data that goes stale faster than a rival can copy it.

  2. Every marketplace is a link prediction problem

    Recommendation, matching, fraud and churn are mostly one question: which connections form next. Treating the business as a graph changes what you can learn.

  3. AI models don’t have network effects. AI products can.

    Foundation models have huge economies of scale and almost no network effects, and switching between them is cheap. The networks in AI live elsewhere.

Part IV Industries & Value

Network effects outside the usual tech examples, and how to find them in the numbers.

  1. UPI and the network effect nobody owns

    UPI beat India’s wallets because its network effect went to a shared rail, not to any app. The same design explains the app concentration and the bill now due.

  2. Industries that don’t call themselves platforms

    Hospitals and insurers, stock exchanges, malls and mandis all run on network effects that nobody names. Each one shows a different part of how they work.

  3. Where network effects show up in the numbers

    A real network effect leaves fingerprints in old cohorts, per-cell outcomes and unit economics, and something else can fake almost every one of them.