Everyone Talks About Network Effects, But Most Companies Don't Actually Have Them

Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle

Network effects are probably the most cited and most misused concept in venture capital circles. Write “we have network effects” in pitch decks and it’s as if a moat magically builds itself. But looking at the major tech companies that emerged over the past thirty years, only a minority truly stand on network effects. For most companies, what they call “network effects” is really just faster growth.

Two Most Common Confusions

The first confusion is treating network effects and viral growth as the same thing. They point in opposite directions: viral growth solves customer acquisition—it’s an offensive move. Network effects build defense—the more users you have, the harder it is for each user to leave. Burn money on subsidies to attract users, and when the buzz fades, everyone leaves. That’s not network effects, that’s buying traffic. The test is simple: assume you stop all user acquisition today. Does your product’s value to a brand new user still rise with the presence of existing users? If yes, you’re getting warm.

The second confusion is equating more data with network effects. More users do generate more data, but for data to constitute network effects, it must form a closed loop: “more usage → more useful data → better product → more usage.” A business with data accumulation but no corresponding usage growth has economies of scale, not network effects. Many companies sit on massive datasets while product experience doesn’t improve—they’ve fallen into exactly this trap.

There’s also the time lag that’s easily overlooked. Network effects don’t appear at launch. They need to cross a critical threshold before self-reinforcement kicks in. Before that threshold, a product’s appeal to new users comes almost entirely from features and experience itself. Claiming network effects in pitch decks at this stage means betting on a future that hasn’t materialized yet. Understanding how far you are from the threshold is far more honest than repeatedly claiming “we have network effects.”

Strength Depends on Why Users Can’t Leave You

Network effects come from different sources and vary wildly in strength. Setting aside specific labels, value sources can be grouped into four types.

Connection-based: Value comes from how many people you can reach through the product. Telephone, instant messaging, professional networking all fall here. This is the strongest type because the cost of leaving is losing an entire relationship graph. The flip side: cold start is hardest—you must first prove there are people on the other end of the network.

Marketplace: Value comes from mutual reinforcement between supply and demand sides. Each additional seller gives buyers more choice; each additional buyer gives sellers more business. This type launches relatively easily, but has a clear weakness: users can actively participate in multiple marketplaces without penalty. When both sides are “multi-tenant,” platform defense is actually quite fragile.

Data-based: Value comes from data users contribute and other users consume. Traffic navigation and restaurant reviews are典型 examples. This type often has a ceiling—after a certain number of reviews, a few more make no difference. Unless the data has strong time sensitivity requiring continuous refresh, the ceiling can’t be broken.

Psychological: Value comes from social psychology. “Not using a certain collaboration tool means you’re not a modern team,” “carrying a certain laptop brand marks you as an insider”—these rely on conformity and identity. It launches fast and shows results quickly, but is the least stable. When the buzz passes, it dissipates. Early adopters may even actively leave when the product becomes mainstream. It works as an ignition source, not a long-term defense line.

The weaknesses of the four types can be viewed side by side:

The Right Question Is “Where’s the Weakness?”

Most founders ask “does my product have network effects?” That question is too coarse. What you should ask are three consecutive sub-questions: Does my value come from connection, marketplace, data, or psychology? What does this type of effect fear most? What have I done in product design to address this weakness?

For example, the biggest vulnerability of marketplace type is multi-tenancy. The core move is to give the supply side value they can’t get elsewhere or create lock-in, so sellers have no incentive to duplicate listings on competitor platforms. Or with psychological type, which inevitably decays, you need to leverage the heat window it ignites to quickly settle users into connection-based or data-based structures.

Using this framework to look at ridesharing: supply-side growth’s value to passengers starts steep then flattens. Wait time dropping from eight minutes to four is night and day; from four to two barely registers. Users on the same demand side can even hurt each other—surge pricing is proof. So this type of marketplace has weaker network effects than imagined. The competitive window stays open long-term. Any new entrant who captures enough supply in a specific region can peel off market share.

Another counterexample worth remembering: psychological effects can inflate beyond the product’s actual value. External valuations of certain star collaboration tools once clearly ran ahead of product utility. Status propped up by buzz will eventually return to the product itself.

The lowest-cost verification move is to ground these three sub-questions in specific design: find a metric that proves value increases with users, then find an action that makes old users unable to leave, and put them in your roadmap. Do these two things and you’re actually building network effects. If you can’t, delete the phrase from pitch decks and focus on solidifying product and retention first.

One final boundary reminder: network effects stories mostly come from winners—natural survivorship bias. The flip side of winner-take-all is the mass of companies claiming network effects that silently died. Network effects are an amplifier. They amplify structure, not growth. Without structure, no amount of growth is anything but bloat.