The Cold Start Problem: Building Networks from Zero

Network effects drive the most successful tech products, but new platforms face a critical cold start problem: they're worthless without users. Andrew Chen presents a five-part framework—atomic networks, easy vs. hard sides, tipping point, escape velocity, and moats—with case studies from Tinder, Slack, Reddit, Airbnb, and LinkedIn showing how founders manually build dense networks before scaling.

What Are Network Effects

The Telephone Principle

A product connecting people is only valuable when others use it too. Theodore Vail's 1915 quote captures this: a telephone without connections is useless; its value grows with each additional connection. Replace 'telephone' with any network product—dating apps, workplace chat, marketplaces—and the principle holds.

Why Network Products Fail at Launch

Products like Dropbox, Slack, Uber, and Tinder become useless if key users aren't present. When you open Slack and no coworkers are there, you switch to what everyone uses. When Uber has no available cars (a 'zero'), the network breaks. This is the cold start problem: without density, the product dies.

The Paradox of Network Effects

Successful network products grow faster as they scale (positive feedback loop), but the inverse is also true: with zero users, a product has zero value and kills new startups. This creates an asymmetry that founders must solve deliberately.

The Five-Part Framework

Cold Start Problem

The initial phase where a new network product has no users and therefore no value, making it impossible to gain traction through normal means.

Atomic Networks

The minimum number of connected users needed for a product to be valuable varies by product type. Zoom needs 2–3 people; Slack needs 5–10 on the same team; Airbnb and Uber need hundreds of listings/drivers per city. Knowing your atomic network size is critical.

Easy Side vs. Hard Side

Every network has two sides: the easy side (consumers, riders, viewers) who do minimal work, and the hard side (creators, drivers, sellers) whose livelihood depends on the product. The hard side is 5–10x more valuable to acquire but also pickier. Success requires optimizing for the hard side first.

Tipping Point

The moment when you can create networks on demand—building the 2nd, 3rd, 5th, and 10th atomic networks in sequence. Once one network is dense and self-sustaining, you can replicate the playbook in new geographies or segments.

Escape Velocity

The phase where you professionalize growth marketing and amplify network effects through paid acquisition, referral programs, and viral loops. The network is now strong enough to absorb and retain users at scale.

Moats and Ceilings

As products scale, they hit ceilings (overcrowding, spam, misuse) and face new competitors. Building defensibility requires features that solve these problems—moderation, filtering, cross-company connections—and creating switching costs that lock in users and data.

Case Study: Tinder—Atomic Networks and the Hard Side

Perfect Product, Zero Traction

Tinder nailed the product on day one—swiping, matching, messaging. But when they invited friends from their address books, adoption failed because online dating carried stigma and they didn't have enough profiles in any one place. Having all the right features meant nothing without network density.

The Birthday Party Hack

Tinder solved the cold start by sponsoring a birthday party for a popular USC student, requiring attendees to install the app to enter. The next day, 500 of the most desirable, hyperconnected people (Greek system leaders) had the app with matches ready. This proved the atomic network concept and became the playbook for college expansion.

Protecting the Hard Side

Tinder's swiping mechanic lets attractive users (the hard side) control match velocity—they can get 10 matches a week or 500. This prevents the inundation problem that plagued earlier dating apps like Match.com, where attractive women received thousands of unmanageable messages.

Network of Networks Strategy

Tinder didn't build one monolithic network; it built a network of networks—one per college, then per city. Dating is hyperlocal (San Francisco users won't date South Bay users an hour away), so Tinder won college by college, then city by city, using the Greek system's hierarchy as a distribution channel in southern states.

Case Study: Slack—From Failure to Atomic Networks

Glitch Failure, Slack Pivot

Stewart Butterfield's game 'Glitch' raised tens of millions and hired dozens but completely failed in market. Rather than shut down, he pivoted: the team had been using IRC chat internally. He rebuilt it into Slack—searchable, non-technical, configurable—and tested it with his core team of 6–7 people as the first atomic network.

Atomic Network Validation

Slack's founding team knew that if the chat product worked for their small team (the atomic network), they could replicate it with other teams. They didn't launch broadly; they manually onboarded friends' companies one by one, building a waitlist to control density and ensure each team had critical mass.

Why Big Companies Fail at This

Large companies try to launch new network products by blasting their existing massive audience. The problem: new users aren't connected to each other, so the network never coalesces. Slack's approach—small, dense, replicable atomic networks—works because each cohort has time to bond organically.

Case Study: Reddit—Flintstoning and Manual Seeding

The Flintstoning Concept

Reddit's founders Steve Huffman and Alexis Ohanian created fake users and posted content manually to fill the home page. When they forgot to run the script, the page was empty and they panicked. This 'human-in-the-loop' bootstrapping—named after Fred Flintstone's foot-powered car—solved the cold start until real users took over.

Manual to Algorithmic Transition

Reddit started with manual posting, then added scripts to scrape headlines from CNN and Digg, then eventually organic user content took over. The progression from human effort to automation to community-driven content mirrors the journey every network product takes.

Why Flintstoning Works Temporarily

Flintstoning creates the illusion of a live, active community, which attracts real users. Once enough real users join and the network reaches critical mass, you can remove the artificial scaffolding and let the community sustain itself.

Case Study: LinkedIn—Targeting the Middle

The Professional Hierarchy Problem

LinkedIn couldn't start with Bill Gates (too successful, no need for networking) or junior professionals (not enough value to offer). Reid Hoffman identified the middle tier—people like Mark Pincus who'd had some success, were still in the grind, and actively needed to recruit and network.

Manual Seeding and Invite-Only

LinkedIn's team manually went through their address books and invited friends one by one. The product was invite-only for the first week, ensuring that early users were already connected to someone on the platform—solving 20% of the atomic network problem immediately.

Invite Mechanics and Network Density

Invites are powerful because they pre-connect new users to at least one existing user. If an atomic network requires 5 connections, every invited user starts 20% of the way there. Invites also copy atomic networks across communities (professionals, alumni, companies), enabling viral saturation within niches.

Case Study: Airbnb—Competing Against Wimdu

The Wimdu Threat

Wimdu, funded with $90 million by Rocket Internet, entered Europe with more funding, more staff, and more inventory than Airbnb. By Metcalfe's law (bigger network wins), Wimdu should have won. Instead, Airbnb dominated because network quality—not just size—matters.

Global Network Effects

Airbnb's US travelers wanted to book in Europe, so the company leveraged its existing network to expand globally. Wimdu had no such pre-existing user base. Quality and density of connections mattered more than raw size, and Wimdu eventually failed.

Why Uber Couldn't Replicate This

Unlike Airbnb, Uber has no global network effects. Success in San Francisco doesn't help in London or Paris—each city is a separate network. This is why Uber could win one city but lose an adjacent one, making it a much harder business than travel platforms with global networks.

Airbnb's Cold Start Solutions

Airbnb solved the cold start by targeting conferences (SXSW) to get guests and hosts at the same time, using Craigslist to post listings and grow on an existing platform, and having early employees use the product themselves (Flintstoning).

Playbook: Do Things That Don't Scale

Manual User Recruitment

Paul Graham's essay 'Do Things That Don't Scale' captures the hard truth: 'The most common unscalable thing founders have to do at the start is to recruit users manually. Nearly all startups have to. You can't wait for users to come to you. You have to go out and get them.' This means throwing parties, cold outreach, and personal invitations—unglamorous but essential.

Invite-Only Mechanics

Invites work for multiple reasons: they ensure infrastructure readiness, create FOMO and status signaling, and pre-connect new users to existing ones. Clubhouse gave invites to the top 25% most active users weekly, ensuring high-quality user acquisition and network density.

Come for the Tool, Stay for the Network

Products like Workspace (Google Docs) started without network effects but evolved into collaborative tools. The lesson: ask whether a non-network product could become network-driven. Hotels didn't think they could be disrupted until Airbnb reimagined lodging as a peer-to-peer network.

Web3 and Network Effects

Cryptocurrencies as Network Effects

Bitcoin is valuable because everyone believes it's valuable—a self-referential network effect. You could fork Bitcoin's code and create Fitcoin, but it would be worthless without a community. This teaches founders that Web3 requires solving the cold start problem just like Web2.

NFTs and Community Building

An NFT project's value depends on solving the cold start: building Discord groups, subreddits, and Twitter communities around the collection. Without network density, the vast majority of NFT projects fail—just like art, where the Mona Lisa's fame is self-referential, not necessarily tied to objective quality.

Web3 Games and Network Lock-In

Games like Axie Infinity and Zed Run combine utility (fun gameplay) with network effects (friends playing together). Games already have massive network lock-in—everyone plays Call of Duty or Fortnite, not new competitors—and Web3 games must overcome this same cold start problem despite blockchain advantages.

Decentralization Requires Network Thinking

Web3's decentralized nature means you can't rely on a central authority to solve problems. You must think deeply about incentives and how the network wants to interact over time, making network effects thinking even more critical than in Web2.

Key Takeaways for Founders

Start in Niches, Not Broadly

Snapchat in high schools, Tinder in colleges, Slack in small teams—all successful networks started in tight niches where they could achieve atomic network density. Broad launches dilute the network and kill adoption.

Know Your Atomic Network Size

Before launching, determine the minimum number of connected users needed for your product to be valuable. This number dictates your go-to-market strategy and helps you avoid the trap of having 'lots of users who aren't connected to each other.'

Optimize for the Hard Side First

Identify which side of your network is harder to acquire and more valuable. Focus your early efforts there. Drivers for Uber, creators for YouTube, sellers for eBay—get them happy first, and the easy side will follow.

Build Density Before Scale

The temptation to go broad early is strong, but it kills network effects. Build one atomic network until it's dense and self-sustaining, then replicate the playbook. This is why Slack went team-by-team and Tinder went school-by-school.

Be Willing to Flintstones

Manual seeding, fake users, founder involvement—these unscalable tactics are temporary scaffolding. Use them to bootstrap network density, then remove them once organic growth takes over. Reddit's fake users and Airbnb's founder stays are examples.

Notable quotes

A telephone without a connection at the other end of the line is not even a toy. Its value depends on the connection with the other telephone and increases with the number of connections. — Theodore Vail, AT&T chairman (1915)
The most common unscalable thing founders have to do at the start is to recruit users manually. Nearly all startups have to. You can't wait for users to come to you. — Paul Graham, Y Combinator
You have to believe that the future is going to be better. We're overexposed to negative things, but the amount of progress in the last couple of decades has been amazing. — Andrew Chen

Action items

  • Define your product's atomic network size—the minimum number of connected users needed for value. Test this hypothesis with early users.
  • Identify the hard side and easy side of your network. Prioritize acquisition and retention of the hard side first.
  • Plan a niche launch strategy targeting one tight community (college, company, neighborhood) where you can achieve atomic network density before expanding.
  • Design invite mechanics or manual onboarding processes to pre-connect new users to existing ones, reducing cold start friction.
  • Accept unscalable tactics early (manual outreach, founder involvement, Flintstoning) as temporary scaffolding to bootstrap network effects.
  • Build features that protect and empower your hard-side users (e.g., Tinder's swipe control for attractive users, Slack's team-based organization).
  • Plan your network-of-networks expansion strategy: how will you replicate success from one atomic network to the next?
Glimpse.
Talks at Google
1 hr video
3 min read
The Cold Start Problem: Building Networks from Zero
You just saved 57 min.
The big takeaway
Network effects drive the most successful tech products, but new platforms face a critical cold start problem: they're worthless without users. Andrew Chen presents a five-part framework—atomic networks, easy vs. hard sides, tipping point, escape velocity, and moats—with case studies from Tinder, Slack, Reddit, Airbnb, and LinkedIn showing how founders manually build dense networks before scaling.
What Are Network Effects
The Telephone Principle
A product connecting people is only valuable when others use it too. Theodore Vail's 1915 quote captures this: a telephone without connections is useless; its value grows with each additional connection. Replace 'telephone' with any network product—dating apps, workplace chat, marketplaces—and the principle holds.
Why Network Products Fail at Launch
Products like Dropbox, Slack, Uber, and Tinder become useless if key users aren't present. When you open Slack and no coworkers are there, you switch to what everyone uses. When Uber has no available cars (a 'zero'), the network breaks. This is the cold start problem: without density, the product dies.
The Paradox of Network Effects
Successful network products grow faster as they scale (positive feedback loop), but the inverse is also true: with zero users, a product has zero value and kills new startups. This creates an asymmetry that founders must solve deliberately.
The Five-Part Framework
Cold Start Problem
The initial phase where a new network product has no users and therefore no value, making it impossible to gain traction through normal means.
Atomic Networks
The minimum number of connected users needed for a product to be valuable varies by product type. Zoom needs 2–3 people; Slack needs 5–10 on the same team; Airbnb and Uber need hundreds of listings/drivers per city. Knowing your atomic network size is critical.
1
Zoom
2–3 people
2
Slack
5–10 team members
3
Tinder
~500 profiles per geography
4
Airbnb/Uber
Hundreds per city
Atomic network size by product type
Easy Side vs. Hard Side
Every network has two sides: the easy side (consumers, riders, viewers) who do minimal work, and the hard side (creators, drivers, sellers) whose livelihood depends on the product. The hard side is 5–10x more valuable to acquire but also pickier. Success requires optimizing for the hard side first.
Cost to acquire driver (Uber)
300 USD
Cost to acquire rider (Uber)
20 USD
Hard side (drivers) costs 10–15x more to acquire than easy side (riders)
Tipping Point
The moment when you can create networks on demand—building the 2nd, 3rd, 5th, and 10th atomic networks in sequence. Once one network is dense and self-sustaining, you can replicate the playbook in new geographies or segments.
Escape Velocity
The phase where you professionalize growth marketing and amplify network effects through paid acquisition, referral programs, and viral loops. The network is now strong enough to absorb and retain users at scale.
Moats and Ceilings
As products scale, they hit ceilings (overcrowding, spam, misuse) and face new competitors. Building defensibility requires features that solve these problems—moderation, filtering, cross-company connections—and creating switching costs that lock in users and data.
Case Study: Tinder—Atomic Networks and the Hard Side
Perfect Product, Zero Traction
Tinder nailed the product on day one—swiping, matching, messaging. But when they invited friends from their address books, adoption failed because online dating carried stigma and they didn't have enough profiles in any one place. Having all the right features meant nothing without network density.
The Birthday Party Hack
Tinder solved the cold start by sponsoring a birthday party for a popular USC student, requiring attendees to install the app to enter. The next day, 500 of the most desirable, hyperconnected people (Greek system leaders) had the app with matches ready. This proved the atomic network concept and became the playbook for college expansion.
Protecting the Hard Side
Tinder's swiping mechanic lets attractive users (the hard side) control match velocity—they can get 10 matches a week or 500. This prevents the inundation problem that plagued earlier dating apps like Match.com, where attractive women received thousands of unmanageable messages.
Network of Networks Strategy
Tinder didn't build one monolithic network; it built a network of networks—one per college, then per city. Dating is hyperlocal (San Francisco users won't date South Bay users an hour away), so Tinder won college by college, then city by city, using the Greek system's hierarchy as a distribution channel in southern states.
Case Study: Slack—From Failure to Atomic Networks
Glitch Failure, Slack Pivot
Stewart Butterfield's game 'Glitch' raised tens of millions and hired dozens but completely failed in market. Rather than shut down, he pivoted: the team had been using IRC chat internally. He rebuilt it into Slack—searchable, non-technical, configurable—and tested it with his core team of 6–7 people as the first atomic network.
Atomic Network Validation
Slack's founding team knew that if the chat product worked for their small team (the atomic network), they could replicate it with other teams. They didn't launch broadly; they manually onboarded friends' companies one by one, building a waitlist to control density and ensure each team had critical mass.
Why Big Companies Fail at This
Large companies try to launch new network products by blasting their existing massive audience. The problem: new users aren't connected to each other, so the network never coalesces. Slack's approach—small, dense, replicable atomic networks—works because each cohort has time to bond organically.
Case Study: Reddit—Flintstoning and Manual Seeding
The Flintstoning Concept
Reddit's founders Steve Huffman and Alexis Ohanian created fake users and posted content manually to fill the home page. When they forgot to run the script, the page was empty and they panicked. This 'human-in-the-loop' bootstrapping—named after Fred Flintstone's foot-powered car—solved the cold start until real users took over.
Manual to Algorithmic Transition
Reddit started with manual posting, then added scripts to scrape headlines from CNN and Digg, then eventually organic user content took over. The progression from human effort to automation to community-driven content mirrors the journey every network product takes.
1
Manual posting by founders
2
Scripts scraping external sites (CNN, Digg)
3
Organic user-generated content dominates
4
Community self-sustains
Reddit's evolution from manual seeding to organic network
Why Flintstoning Works Temporarily
Flintstoning creates the illusion of a live, active community, which attracts real users. Once enough real users join and the network reaches critical mass, you can remove the artificial scaffolding and let the community sustain itself.
Case Study: LinkedIn—Targeting the Middle
The Professional Hierarchy Problem
LinkedIn couldn't start with Bill Gates (too successful, no need for networking) or junior professionals (not enough value to offer). Reid Hoffman identified the middle tier—people like Mark Pincus who'd had some success, were still in the grind, and actively needed to recruit and network.
Manual Seeding and Invite-Only
LinkedIn's team manually went through their address books and invited friends one by one. The product was invite-only for the first week, ensuring that early users were already connected to someone on the platform—solving 20% of the atomic network problem immediately.
Invite Mechanics and Network Density
Invites are powerful because they pre-connect new users to at least one existing user. If an atomic network requires 5 connections, every invited user starts 20% of the way there. Invites also copy atomic networks across communities (professionals, alumni, companies), enabling viral saturation within niches.
Case Study: Airbnb—Competing Against Wimdu
The Wimdu Threat
Wimdu, funded with $90 million by Rocket Internet, entered Europe with more funding, more staff, and more inventory than Airbnb. By Metcalfe's law (bigger network wins), Wimdu should have won. Instead, Airbnb dominated because network quality—not just size—matters.
Wimdu funding
90 million USD
Airbnb funding (at time)
50 million USD (approx)
Wimdu had more capital but lost to Airbnb's network quality
Global Network Effects
Airbnb's US travelers wanted to book in Europe, so the company leveraged its existing network to expand globally. Wimdu had no such pre-existing user base. Quality and density of connections mattered more than raw size, and Wimdu eventually failed.
Why Uber Couldn't Replicate This
Unlike Airbnb, Uber has no global network effects. Success in San Francisco doesn't help in London or Paris—each city is a separate network. This is why Uber could win one city but lose an adjacent one, making it a much harder business than travel platforms with global networks.
Airbnb's Cold Start Solutions
Airbnb solved the cold start by targeting conferences (SXSW) to get guests and hosts at the same time, using Craigslist to post listings and grow on an existing platform, and having early employees use the product themselves (Flintstoning).
Playbook: Do Things That Don't Scale
Manual User Recruitment
Paul Graham's essay 'Do Things That Don't Scale' captures the hard truth: 'The most common unscalable thing founders have to do at the start is to recruit users manually. Nearly all startups have to. You can't wait for users to come to you. You have to go out and get them.' This means throwing parties, cold outreach, and personal invitations—unglamorous but essential.
Invite-Only Mechanics
Invites work for multiple reasons: they ensure infrastructure readiness, create FOMO and status signaling, and pre-connect new users to existing ones. Clubhouse gave invites to the top 25% most active users weekly, ensuring high-quality user acquisition and network density.
Come for the Tool, Stay for the Network
Products like Workspace (Google Docs) started without network effects but evolved into collaborative tools. The lesson: ask whether a non-network product could become network-driven. Hotels didn't think they could be disrupted until Airbnb reimagined lodging as a peer-to-peer network.
Web3 and Network Effects
Cryptocurrencies as Network Effects
Bitcoin is valuable because everyone believes it's valuable—a self-referential network effect. You could fork Bitcoin's code and create Fitcoin, but it would be worthless without a community. This teaches founders that Web3 requires solving the cold start problem just like Web2.
NFTs and Community Building
An NFT project's value depends on solving the cold start: building Discord groups, subreddits, and Twitter communities around the collection. Without network density, the vast majority of NFT projects fail—just like art, where the Mona Lisa's fame is self-referential, not necessarily tied to objective quality.
Web3 Games and Network Lock-In
Games like Axie Infinity and Zed Run combine utility (fun gameplay) with network effects (friends playing together). Games already have massive network lock-in—everyone plays Call of Duty or Fortnite, not new competitors—and Web3 games must overcome this same cold start problem despite blockchain advantages.
Decentralization Requires Network Thinking
Web3's decentralized nature means you can't rely on a central authority to solve problems. You must think deeply about incentives and how the network wants to interact over time, making network effects thinking even more critical than in Web2.
Key Takeaways for Founders
Start in Niches, Not Broadly
Snapchat in high schools, Tinder in colleges, Slack in small teams—all successful networks started in tight niches where they could achieve atomic network density. Broad launches dilute the network and kill adoption.
Know Your Atomic Network Size
Before launching, determine the minimum number of connected users needed for your product to be valuable. This number dictates your go-to-market strategy and helps you avoid the trap of having 'lots of users who aren't connected to each other.'
Optimize for the Hard Side First
Identify which side of your network is harder to acquire and more valuable. Focus your early efforts there. Drivers for Uber, creators for YouTube, sellers for eBay—get them happy first, and the easy side will follow.
Build Density Before Scale
The temptation to go broad early is strong, but it kills network effects. Build one atomic network until it's dense and self-sustaining, then replicate the playbook. This is why Slack went team-by-team and Tinder went school-by-school.
Be Willing to Flintstones
Manual seeding, fake users, founder involvement—these unscalable tactics are temporary scaffolding. Use them to bootstrap network density, then remove them once organic growth takes over. Reddit's fake users and Airbnb's founder stays are examples.
Worth quoting
"A telephone without a connection at the other end of the line is not even a toy. Its value depends on the connection with the other telephone and increases with the number of connections."
— Theodore Vail, AT&T chairman (1915), at [4:42]
"The most common unscalable thing founders have to do at the start is to recruit users manually. Nearly all startups have to. You can't wait for users to come to you."
— Paul Graham, Y Combinator, at [30:26]
"You have to believe that the future is going to be better. We're overexposed to negative things, but the amount of progress in the last couple of decades has been amazing."
— Andrew Chen, at [58:00]
Try this
Define your product's atomic network size—the minimum number of connected users needed for value. Test this hypothesis with early users.
Identify the hard side and easy side of your network. Prioritize acquisition and retention of the hard side first.
Plan a niche launch strategy targeting one tight community (college, company, neighborhood) where you can achieve atomic network density before expanding.
Design invite mechanics or manual onboarding processes to pre-connect new users to existing ones, reducing cold start friction.
Accept unscalable tactics early (manual outreach, founder involvement, Flintstoning) as temporary scaffolding to bootstrap network effects.
Build features that protect and empower your hard-side users (e.g., Tinder's swipe control for attractive users, Slack's team-based organization).
Plan your network-of-networks expansion strategy: how will you replicate success from one atomic network to the next?
Made with Glimpse by Wozart
glimpse.wozart.com/v/3r1vtcph
Share this infographic

More like this