Efek Jaringan: Siapa yang Diuntungkan Peserta Berikutnya?
A network effect occurs when a product's value to one user changes as other relevant users join or participate. That change can be positive or negative, and it can run within one group of users or across groups, such as buyers and sellers. Here's the detail that stops most marketers. Dropbox's IPO filing credited growth to viral, bottom-up adoption and, separately, described network effects from sharing as increasing usefulness and stickiness. The company treated them as two different engines, while many growth decks still treat them as one.
The difference matters because only one of those engines makes the product better for the people already using it. This entry covers how the mechanism works, what it gets confused with, where it stops working, and a five-question test for checking a claim before you build strategy on it.
How a Network Effect Actually Works
The idea comes from economics, not growth marketing. In their 1994 paper on systems competition, Katz and Shapiro describe markets where the value a buyer gets depends on how many others use the same or compatible products, with expectations about future adoption shaping which system people choose. Notice the word compatible. Two people on incompatible systems don't benefit from each other at all.
Two qualifiers make the definition useful in practice. First, the other users have to be relevant: a new user in another city, category, or language may add nothing to your experience. Second, the change isn't always upward. More participants can mean more choice for one group and more crowding or competition for another.
The academic sources here are old, and that's fine. They explain a mechanism rather than report a current benchmark, so their age doesn't make them stale.
Direct, Cross-Side, and Negative Effects
Direct Network Effects
A direct effect works within one group of users. A messaging or collaboration tool gets more useful when the specific people you need to reach are on it. The key word is specific. Accounts you'll never interact with don't count.
Cross-Side Network Effects
A cross-side effect runs between groups. Parker and Van Alstyne's 2005 model of two-sided markets shows how participation on one side can raise value for the other, which helps explain why a platform may price one side low to attract the side that pays. In ecommerce, that's the buyer and seller relationship.
The OECD's Digital Economy Outlook adds a condition worth keeping. Buyers gain from more sellers when those sellers widen the scope or variety of products on offer, and sellers gain from a large pool of potential buyers. A marketplace that adds many near-identical listings hasn't necessarily given buyers anything new.
Negative Effects and Congestion
Cross-side and same-side effects can point in opposite directions at the same moment. A new seller may help buyers while competing with every existing seller in the same category for the same attention. That's a negative same-side effect sitting right next to a positive cross-side one.
Congestion needs a sharper diagnosis than blaming crowd size. Ask what the actual constraint is. Limited capacity, such as the drivers available in one area at one hour, is a different problem from competition among participants, even when both show up as worse outcomes for someone on the platform.
What a Network Effect Is Not
Three mechanisms get labeled as network effects all the time. Each can coexist with one. None of them proves one exists.
Viral Growth
Viral growth is about how adoption spreads from person to person. NFX separates the two cleanly, arguing that viral effects can exist without network effects and that the reverse is also true. A referral loop can fill a product with users who never interact. CRV's test is blunt: if users aren't interacting with each other through the product, it isn't a network effect, however fast it grows. Programs built on how viral marketing spreads adoption answer an acquisition question, not a value question.
Supply-Side Economies of Scale
Scale can lower a company's cost per unit. That's a supply-side benefit, and it can make prices lower for everyone. It isn't the same as a user getting more value because other users joined. A business can have both, and the distinction matters when you're explaining why a leader stays ahead. Cheaper fulfillment from volume and wider choice from more sellers are separate stories.
Data Scale and AI Learning
Data flywheels deserve the most skepticism. Hagiu and Wright, in a December 2021 manuscript on data-enabled learning, separate learning across users from learning within a single user. Learning from one person's repeat behavior can improve the product for that same person, which is personalization, not a network effect. Learning across users can support one, but only when there's an explicit path by which data from some users improves outcomes for others. CRV makes a related point: many claimed data network effects are really data scale effects, and a single user working with an AI tool doesn't create a network effect on its own.
Agents raise an open question rather than a settled answer. Andrew Chen argued in October 2026 that agents don't automatically have network effects, since interoperable agents could stay fragmented the way email works across providers, and that the real question is where any network effect ends up living. That's a practitioner view, not a consensus.
Network Effect Mechanisms: Who Benefits and What to Verify
| Network Effect Mechanism | Who Benefits | What to Verify |
|---|---|---|
| Direct effect | Users in the same group who can reach each other | Whether the specific people you need are on it |
| Cross-side effect | Buyers gain from varied sellers; sellers gain from many buyers | Whether new sellers add variety, not near-identical listings |
| Viral growth | Acquisition, not necessarily existing users | Whether users interact with each other through the product |
| Supply-side economies of scale | Customers, through lower prices | Whether value comes from cost, not other users joining |
| Cross-user data learning | Other users, if data from some improves outcomes for others | Whether an explicit path exists, or it is only personalization |
Why Relevant Density Beats Total Users
Local, Time, and Category Density
A global user count says little about the network any one person can use. A rider waiting downtown can't be picked up by a driver in another country. A buyer hunting for one niche category gets nothing from sellers in a different one. Uber's 2019 IPO filing framed its strategy around building the largest network in each market, not the largest network overall. That framing is the company's own, but the practical point holds: density should be measured where interactions actually happen.
Density doesn't automatically lock participants in, either. A 2025 Centre for Competition Policy study of food delivery platforms in the Philippines found that higher demand density was associated with more rider multihoming, meaning riders working across several platforms. The study relied on cross-sectional surveys, so it shows an association, not a cause.
Cold Start and Seeding One Side
Before a network is useful, it's often useless. Two-sided market theory explains why platforms discount or subsidize one side to bring the other in, and Uber's 2019 filing said incentives could produce a negative margin until it reached enough scale. Uber also listed failing to keep a critical mass of drivers and consumers as a risk. Worth checking in any growth story: subsidized growth isn't evidence of an organic network effect. Design matters too, because participants can only add value if the product lets them find each other, which makes it a web design and development problem as much as a marketing one.
When Network Effects Don't Apply or Backfire
Network effects are conditional. They weaken, disappear, or turn negative under conditions you can usually spot in advance.
- Participants don't interact. If users only consume from the company and never benefit from each other, growth is just growth.
- The relevant network is thin. Strong national numbers can hide weak local or category coverage.
- Multihoming is cheap. When buyers, sellers, or drivers can use several platforms at once, a bigger network stops working as a lock. Uber's filings list low switching costs as a risk.
- Size erodes differentiation. In a study of a pet-sitting platform merger, Farronato, Fong, and Fradkin found that the benefits of a larger combined network were offset by the loss of differentiation between the platforms. That result applies to the market it studied, not to every merger.
- More supply means more crowding. Extra sellers can bury good listings and squeeze existing sellers.
The OECD adds a related caution, noting that switching costs and winner-take-most outcomes appear in some digital markets rather than all of them. Treating a network effect as a guaranteed moat is the most expensive mistake on this list. Rising customer retention as a metric alongside user growth looks encouraging, but it can also come from product improvements, discounts, or a company mandating a tool.
Two Company Examples From Public Filings
Both examples come from the companies' own securities filings. They describe what management believed, not independently measured effects.
Uber: A Company-Described Liquidity Loop
Uber's 2019 Form S-1 described a liquidity network effect as a loop. More drivers lead to shorter wait times and lower fares, which draw more riders, which raise rides per hour and driver earnings, which attract more drivers. The company said it believed the operator with the larger network in a given market would earn a margin advantage, and it described its two-way rating system as a way to increase accountability among participants. The same filing listed low barriers to entry and low switching costs among its risks. Its annual report for fiscal 2025 still points to low switching costs and intense competition, and cautions that scale and liquidity may not produce a lasting competitive advantage. Read together, the two documents describe a mechanism the company relies on and limits it openly acknowledges.
Dropbox: Sharing Network and Viral Growth
Dropbox's 2018 Form S-1 described a global sharing network and said that as more users share and collaborate, network effects increase the platform's usefulness and stickiness. In the same filing, Dropbox credited viral, bottom-up adoption, with users bringing the product into their workplaces. Those are two claims about two mechanisms. Recruiting a colleague is acquisition. Collaborating with that colleague afterward is where any network value would come from. The filing also bundled sharing, product insights, and machine learning into a single flywheel, which is exactly the kind of bundle worth unpacking before anyone calls it a network effect.
The Next-Participant Test
This is an editorial checklist, not a validated causal test. Use it to question a network effect claim, then look for evidence that rules out other explanations. Picture the next person who joins and ask five questions.
- Who benefits? Name the existing users whose experience this person could improve, and anyone it could make worse.
- Relevant market? Can those users actually interact with the newcomer in the same location, category, or time window?
- Useful contribution? Does the newcomer add real availability, variety, or quality, or just another account?
- Capacity or quality? Could this arrival create congestion, crowding, or a drop in quality for someone else?
- Alternative platforms? Is the newcomer also active elsewhere, so the value isn't exclusive to you?
If you can't answer the first two, you probably have a growth story. Stronger claims need a research design that accounts for who chose to join and for spillovers between users, which is far more than a dashboard showing users and revenue rising together. For marketplaces, the test pairs naturally with internet marketing channel planning, because acquisition and network value need separate metrics.
Marketing can help a product reach its first users, but it does not create a network effect. That comes from how the product connects the people using it. For early acquisition on a marketplace or platform, teams sometimes use social media ads or influencer marketing to reach the first group, and ecommerce SEO for product and category pages. Whether any of that leads to value for the next participant still has to be checked against the test above.
Frequently Asked Questions
What is a network effect?
A network effect occurs when a product's value to one user changes as other relevant users join or participate. It can be positive or negative and can work within one user group or across groups, such as buyers and sellers.
What's the difference between direct and cross-side network effects?
A direct effect works within one group, such as colleagues on a collaboration tool. A cross-side effect works between groups, such as buyers and sellers on a marketplace. The same platform can have both, and they can point in different directions.
Is a network effect the same as viral growth?
No. Viral growth describes how people adopt a product through others. A network effect describes how other users' participation changes the product's value. A product can spread virally while existing users gain nothing from new ones.
Can network effects be negative?
Yes. More participants can mean congestion, longer waits, or tougher competition. On a marketplace, a new seller can help buyers while hurting other sellers in the same category.
Do network effects guarantee a winner-take-all market?
No. Cheap multihoming, differentiation, and low switching costs can keep several platforms viable. Even Uber's own filings list low switching costs as a risk to its position.
Does an AI data flywheel create a network effect?
Not automatically. If data from one user only improves that user's experience, it's personalization. A network effect needs a clear path by which data or participation from some users improves outcomes for others.
Related Terms and Sources
Related Terms
- Viral marketing and adoption: how products spread from person to person.
- Referral marketing for acquisition: incentive-driven acquisition, not proof of network value.
- Customer retention as a metric: a possible outcome that needs careful attribution.
- User-generated content (UGC) explained: what participants contribute to each other's experience.
Sources
- Systems Competition and Network Effects, Michael L. Katz and Carl Shapiro, Journal of Economic Perspectives, 1994.
- Two-Sided Network Effects theory paper, Geoffrey G. Parker and Marshall W. Van Alstyne, Management Science, 2005.
- OECD Digital Economy Outlook, business models, chapter on evolving business models, OECD, 2020.
- Viral Effects vs. Network Effects, NFX.
- Uber Form S-1 registration statement, Uber Technologies, Inc., filed with the SEC in 2019.
- Uber annual report, fiscal 2025, Form 10-K, Uber Technologies, Inc.
- Dropbox Form S-1 registration statement, Dropbox, Inc., filed with the SEC in 2018.
- Data-enabled learning and network effects, Andrei Hagiu and Julian Wright, author manuscript dated December 2021, not the final published version.
- Dog Eat Dog platform merger study, Chiara Farronato, Jessica Fong, and Andrey Fradkin, 2024.
- Multihoming in food delivery platforms, Centre for Competition Policy, 2025.
- Will agents have network effects?, Andrew Chen, October 2026.
- CRV guide to network effects, CRV, 2026.
