Network effects: types, traps, and how to test for them.

Network effects are the most-claimed and least-understood category of defensibility. Half the pitch decks in any given week claim some form of network effect, and most of those claims do not survive scrutiny. The fix starts with understanding that "network effect" is not one thing. It is at least seven distinct mechanisms, each with different evidence requirements, different scaling dynamics, and different traps.

Direct network effects exist when each additional user increases the value of the product for other users of the same type. The canonical example is a communications platform: each new user on WhatsApp makes WhatsApp more useful for everyone else. The dynamic is "more users equals more value to existing users."

The test for a real direct network effect is whether the product would become meaningfully less useful if a significant percentage of users left. If WhatsApp lost half its users, the value to remaining users would drop substantially because they could communicate with fewer people. If a workflow tool lost half its users, the remaining users would still get the same value because their work does not depend on the other users.

Direct network effects are rare in B2B SaaS because most business tools do not require cross-company coordination. They are common in consumer platforms (social, messaging) and in coordination products (calendaring, video conferencing).

Two-sided network effects exist in marketplaces where more participants on one side attract more participants on the other side, who attract more participants back. Uber needs drivers to attract riders, and riders to attract drivers. eBay needs sellers to attract buyers, and buyers to attract sellers.

The test is whether the marketplace dynamics actually pull each side. A marketplace where one side has to be aggressively recruited at all stages of growth is not a real two-sided network effect; it is a marketplace where the pull works in one direction but not the other. Most failed marketplaces fail because the pull breaks in one direction, typically the supply side.

Data network effects exist when the product accumulates data from user activity, and that data improves the product's value for all users. Netflix's recommendation engine improves with viewing data. Spotify's discovery improves with listening data. The dynamic compounds because more users means more data means a better product means more users.

The test for a real data network effect is whether the data quality could be reproduced by a well-funded competitor in eighteen months. If yes, the data is a head start rather than a moat. If the data requires user behavior at scale that cannot be purchased or simulated, the moat is real.

Most B2B SaaS products that claim data network effects do not have them. Logging customer activity produces data, but that data does not improve the product for other customers because customer activity is mostly independent across companies. A true data network effect in B2B usually requires cross-customer learning, where data from one customer's use case improves the product for other customers in similar use cases.

Social network effects exist when users derive value from the presence of specific other users, particularly users they already know. LinkedIn is more valuable when your professional contacts are on it. Snapchat is more valuable when your friends are on it. The dynamic depends on the social graph, not just the user count.

The test is whether the product would be less valuable if a random 50 percent of users left versus if a specific 50 percent (the user's actual contacts) left. If the latter is more damaging than the former, the network effect is social.

Personal utility network effects exist when the value of the product to a single user increases with the user's own data accumulation. A note-taking app becomes more valuable as the user accumulates notes. A photo storage product becomes more valuable as the user accumulates photos. The dynamic is single-user rather than multi-user, but it produces a switching cost that resembles a network effect.

These are technically switching costs rather than network effects in the strict sense, but they share the compounding dynamic with use, and many pitch decks describe them as network effects.

Standards-based network effects exist when the product becomes the default convention for a category, such that other products integrate with it because it is the standard. Stripe became a network effect through this mechanism: developers integrate Stripe because it is the convention, and other tools support Stripe because developers integrate it.

The test is whether the product has been adopted as the integration target by third parties without paying them to do so. If yes, the standards-based effect is real. If the product is paying for integrations, it is a partnership program, not a network effect.

Three traps recur in network effect claims. Trap one: confusing growth with network effect. Many products grow because of marketing and product quality, not because of cross-user dependencies. Growth is necessary but not sufficient to demonstrate a network effect. Trap two: confusing community with network effect. A community of users who share interests is not the same as a network where users depend on each other. Trap three: confusing data accumulation with data network effect. Accumulating data is useful, but the dynamic only qualifies as a network effect if the data improves the product for users.

The cleanest test is to measure cohort retention as the user base grows. A real network effect produces improving retention over time: later cohorts retain better because the product is more valuable due to the larger network. If retention is flat or declining as the user base grows, the network effect is weak or absent. According to NFX's Network Effects Bible, products with strong network effects typically show 20 to 40 percent better month-three retention for cohorts acquired at 10x the user base versus cohorts acquired at 1x.

Network effects are real, valuable, and rare. Most B2B SaaS products do not have them. Most consumer products do not have them. The companies that have them tend to have one of seven specific types (direct, two-sided, data, social, personal utility, standards-based, or two-sided variations), and the type matters because each one scales differently and faces different threats. A founder claiming a network effect should be able to name the specific type, show the cross-user dependency, and produce cohort data that demonstrates retention improvement as the network grows. For the broader defensibility view, see five real moats for early-stage startups.