Claude Penland

By Claude Penland - marketing and business strategy for companies that are good at what they do and hard to find.

Most Network Effects Are Not Network Effects

A Q&A field guide to telling the real thing from scale, lock-in and wishful thinking

Sit through enough pitch meetings and you will hear “network effects” used the way restaurants use “artisanal.” It gets attached to anything with more than one customer. This guide sorts the real thing from the look-alikes, with numbers, named sources and tests you can run this week.

1. What is a network effect, in plain English?

A product has one when each new user makes it more valuable for the users already there. Hamilton Helmer’s definition in 7 Powers is the cleanest: it occurs “when the value of a product to a customer is increased by the use of the product by others.” James Currier of NFX turns it into a design rule: “Let customer 2 add value to customer 1.” A phone with one owner is a paperweight. A spreadsheet with one owner is still a perfectly good spreadsheet.

2. Why does nearly every pitch deck claim one?

Because the prize is real. NFX studied 336 digital companies founded between 1994 and 2017 that grew past $1 billion in value, and found:

  • 35% of them had network effects at the core, yet those companies held 68% of the total value.
  • The other 65% (219 companies) leaned on scale, brand or embedding, and their valuations “typically top out in the $1-$2B range.”
  • Only 20% of the business plans NFX’s partners saw before starting the firm had network effects in them.

Read the fine print, though. The same study filed Amazon as “100% NON-network-effect” because its core defensibility is “a scale effect.” If the most network-obsessed firm in venture capital says Amazon runs on scale, your invoicing app can live with the same label. Morningstar’s moat research adds a caution: 24% of its wide-moat companies benefit from a network effect versus under 9% of narrow-moat ones, and across a 10-year history the network-effect group “had the least stability” in earnings.

3. What are the real types?

NFX catalogs 16 kinds in its Network Effects Manual. Six cover most of what you will meet in the wild:

  • Direct: WhatsApp, the telephone. How user 2 helps user 1: One more person you can reach. The catch: Strongest kind, and brutal to start from zero.
  • Marketplace: Airbnb, eBay. How user 2 helps user 1: More sellers draw buyers, who draw more sellers. The catch: Works best when supply is varied (a16z).
  • Platform: iOS, Windows. How user 2 helps user 1: Users attract developers; apps attract users. The catch: Both sides must be fed at once.
  • Payments: Visa. How user 2 helps user 1: Each cardholder makes acceptance worth it for merchants. The catch: Visa, fiscal 2025: nearly 5 billion credentials, 175 million+ merchant locations.
  • Local: Uber, Lyft. How user 2 helps user 1: More drivers mean shorter waits. The catch: Value flattens once waits are short.
  • Data: Waze. How user 2 helps user 1: Each driver’s reports fix everyone’s route. The catch: Rare. Most “data network effects” are scale effects.

4. How much is a bigger network actually worth?

Less than the famous formula says. Metcalfe’s Law holds that value grows with the square of users. In IEEE Spectrum, Bob Briscoe, Andrew Odlyzko and Benjamin Tilly argued the truer curve is n log(n), because connections are unequal in value. They borrowed Thoreau’s 1854 jab at the telegraph: “Maine and Texas, it may be, have nothing important to communicate.” Their worked example starts with 100,000 members worth $1 million:

  • Sarnoff (broadcast). Formula: n. Users double: $2.0 million. Users grow 10x: $10 million.
  • Briscoe, Odlyzko, Tilly. Formula: n log(n). Users double: $2.1 million. Users grow 10x: $12 million.
  • Metcalfe. Formula: n squared. Users double: $4.0 million. Users grow 10x: $100 million.

*10x column is my arithmetic using the same formulas. The gap between the last two rows is roughly the size of the dot-com bubble; the authors note Cisco’s stock fell 89%, more than $580 billion in paper value, between March 2000 and October 2002. In fairness to Metcalfe, a 2015 paper by Zhang, Liu and Xu fit real Facebook and Tencent data and found his law fit best. Squared math can hold for a genuine communication network. A note-taking app is a different animal.

5. How do I test whether mine exists?

Start with what I call the empty-room test, then check five of the measures from a16z’s “16 Ways to Measure Network Effects” by Li Jin and D’Arcy Coolican.

  • Empty room (one user). Looks like the real thing: Useless or far weaker. Looks like an impostor: Works fine solo.
  • Organic signups. Looks like the real thing: Rises over time. Looks like an impostor: Growth tracks ad spend.
  • Acquisition cost. Looks like the real thing: Falls as the network grows. Looks like an impostor: Flat or rising.
  • Retention by cohort. Looks like the real thing: Newer cohorts stick better. Looks like an impostor: Every cohort decays alike.
  • Multi-homing. Looks like the real thing: Users stay on one network. Looks like an impostor: Users run a rival’s app too.
  • Pricing power. Looks like the real thing: Prices rise, users stay. Looks like an impostor: Discounts hold the base.

Real networks also show a threshold where other people start doing the retaining for you. Facebook’s growth team chased “7 friends in 10 days.” Slack’s Stewart Butterfield reported that once a team had exchanged 2,000 messages, “93% of those customers are still using Slack today.” If your activation metric involves zero other humans, take the hint.

6. How is it different from scale economies and switching costs?

All three keep rivals out, which is why they get confused. Helmer names a different barrier for each.

  • Network effect. What happens: Better as users join. The tell: Buyers ask who else is on. A rival must: Rebuild the whole network. Examples: Visa, WhatsApp.
  • Scale economies. What happens: Cheaper per unit at volume. The tell: Margins widen, same product. A rival must: Match your volume. Examples: Amazon retail, Netflix content.
  • Switching costs. What happens: Leaving is costly. The tell: Customers grumble, then renew. A rival must: Pay customers to move. Examples: Salesforce, payroll software.

A fourth impostor is virality. It gets you users cheaply and does little to keep them. Zoom went from 10 million daily meeting participants in December 2019 to 300 million by April 2020, and in that same stretch Microsoft Teams reached 75 million daily active users, up 70% in about six weeks. A meeting link works for anyone on any service, so guests owe the host’s vendor no loyalty. a16z’s warning fits: companies great at viral growth without network effects “can grow quickly but flame out just as fast.”

7. Isn’t my data a network effect?

Probably a scale effect, and a leaky one. In “The Empty Promise of Data Moats,” a16z’s Martin Casado and Peter Lauten show a customer-support chatbot where 20% of the effort bought about 20% coverage and the curve flattened at 40% of customer intents. Past that point more data cost more and added little. Their advice to founders is blunt: “don’t even assume you have a data network effect (you likely don’t).”

8. What do weak and fake network effects cost?

  • Ridesharing (weak). a16z notes that once pickup waits hit about 5 minutes, riders stop caring how many more drivers a platform has. Drivers hedge too: Gridwise data for early 2024 shows about 55% of Lyft drivers also working Uber, and just over 25% of Uber drivers also on Lyft. Uber ran up $31.5 billion in operating losses from 2014 before its first operating profit in 2023. Harvard’s Feng Zhu and Marco Iansiti explained why in HBR: local clustering and multi-homing sap a platform’s strength.
  • WeWork (fake). Its 2019 filing claimed that “the more locations we strategically cluster in a given city, the larger and more dynamic our community becomes.” Greg Satell ran the desk math: rival IWG earned almost $7,000 per desk on about $6,000 of cost, while WeWork took in just over $4,000 against more than $8,000. Private investors valued WeWork at $47 billion; the market valued IWG, with slightly more space, under $4 billion.

9. Can a real network effect fade or reverse?

Yes, and it happens more often than the textbooks suggest. Coolican and Jin describe Frank, a lending app for friends and family: adding friends helped until a group passed 7 people, after which members became less likely to lend or borrow. Trust has a headcount. A rival with an overlapping network can also walk in the side door, as Instagram Stories did to Snapchat. Zhu and Iansiti open their HBR piece with Didi, which reached 25 million trips a day in 2016 after pushing Uber out of China, only to watch Meituan launch ride-hailing in early 2018 just as profits came into view.

10. What should I claim when the honest answer is neither?

Claim what you have. Helmer lists seven sources of durable advantage and five have nothing to do with networks or scale. A specific, checkable claim stands out.

  • Switching costs: “Customers keep four years of records with us and renew at X%.” Morningstar found this group’s returns the most stable.
  • Cornered resource: exclusive data, licenses or contracts. Casado and Lauten cite Equifax and LexisNexis.
  • Brand, process or counter-positioning: a reputation, a way of operating, or a model incumbents can’t copy without hurting themselves.
  • Plain execution: a16z’s list includes domain depth, go-to-market strength and talent. “We sell faster and know this industry cold” is a respectable sentence.
  • A plan: “No moat yet. Here is the sequence that builds one.” Honest, and rarer than you would think.

A quick translation guide for the next board meeting:

  • “Every customer makes our AI smarter”. The honest label: Data scale effect. Find where the curve flattens.
  • “Users invite their coworkers”. The honest label: Virality, unless coworkers improve it for each other.
  • “We have 2 million users”. The honest label: Size. Check retention and multi-homing.
  • “Customers can’t rip us out”. The honest label: Switching costs. Say so proudly.
  • “More volume, lower prices”. The honest label: Scale economies, the Amazon kind.
  • “Buyers come for sellers, sellers for buyers”. The honest label: A marketplace network effect. Now measure it.

11. What is the 30-second version?

Ask whether customer 2 makes the product better for customer 1. If yes, measure it with the table in question 5. If your costs fall, call it scale. If customers can’t leave, call it switching costs. If neither, say so and show the plan. People who know the difference will trust the rest of your deck more for it.

Sources

  1. NFX, “70 Percent of Value in Tech” (2019): nfx.com/post/70-percent-value-network-effects
  2. NFX, “The Network Effects Manual”: nfx.com/post/network-effects-manual
  3. Hamilton Helmer, 7 Powers: The Foundations of Business Strategy (2016).
  4. IEEE Spectrum, “Metcalfe’s Law Is Wrong” (2006): spectrum.ieee.org/metcalfes-law-is-wrong
  5. Zhang, Liu, Xu, Metcalfe validation (2015): link.springer.com/article/10.1007/s11390-015-1518-1
  6. a16z, “16 Ways to Measure Network Effects”: a16z.com/16-ways-to-measure-network-effects/
  7. a16z, “The Dynamics of Network Effects”: a16z.com/the-dynamics-of-network-effects/
  8. a16z, “The Empty Promise of Data Moats”: a16z.com/the-empty-promise-of-data-moats/
  9. Zhu and Iansiti, HBR (2019): hbr.org/2019/01/why-some-platforms-thrive-and-others-dont
  10. Morningstar moat study: morningstar.co.uk/uk/news/69802/companies-that-stay-ahead-of-the-competition.aspx
  11. Visa, Fiscal 2025 Annual Report: annualreport.visa.com
  12. Gridwise, multi-apping data (2024): gridwise.io/blog/analytics/revealing-rideshare-trends-multi-apping-and-airport-insights
  13. Irish Times/FT on Uber (2023): irishtimes.com/business/2023/08/01/uber-makes-first-operating-profit-after-racking-up-315bn-of-losses/
  14. TechRadar on Zoom and Teams (2020): techradar.com/news/zoom-admits-lying-about-user-numbers
  15. Greg Satell on WeWork (2019): digitaltonto.com/2019/the-wework-debacle-is-just-a-symptom-of-a-much-larger-problem
  16. Stewart Butterfield on Slack, quoted in Andrew Chen, The Cold Start Problem (2021).

Discover more from 1000 Startups

Subscribe to get the latest posts sent to your email.

Claude Penland

Claude Penland builds the marketing and business strategy for companies that are good at what they do and hard to find. Thirty years operating, one exit, eight of them as a practicing casualty actuary.

The free two-page read is genuinely free. Email claude@1000startups.com and I'll send back what I can see from the outside. Or see the work samples and how to work with me.

Leave a Reply