Book summary: The Cold Start Problem

Using network effects to scale your product

George Peppou
7 min readJan 17, 2022

1 Paragraph Summary

Network effects describes products that become more valuable the more people use it. From credit cards, to social media, to mobility. Building a company like this happens in stages, the first, the ‘cold start’ is the most challenging to overcome as it requires a network to be built before it can offer the value that people join it for. This book is a bible of the tactics which can be used to overcome this challenge.

Network effects have defined many of the most successful software companies of the past two decades. It’s a tantalisingly simple concept: the more users a network has the more valuable it becomes for us. From Twitter only be interesting if there are other interesting people to follow to Uber relying on lots of riders to utilise lots of drivers to ensure wait times are short network effects have defined business for the past couple of decades.

But how does a network actually form? Chen breaks it down into five steps:

  1. The Cold Start Problem — how do you build a first, small network with enough users to make the product valuable?
  2. Tipping Point — once you have built this first network each subsequent network gets a little easier.
  3. Escape Velocity — once scale is hit quickly three complementary advantages come into play: low cost user acquisition through networks, high engagement as the network is so robust & improved economics as the network grows.
  4. Hitting the Ceiling — something happens which stalls growth, this may be saturation, scams or something else entirely. Products must adjust to respond and to overcome this.
  5. The Moat — a strong, established network becomes a strong moat over time.
See, it’s simple!

Each of these stages are looked into in great detail.

The Cold Start Problem

Unsurprisingly, as the books namesake this first step gets a lot more attention than the other stages.

The atomic network

The atomic network is the first network that the product is adopted by, this is the smallest possible group in which this can provide value. In every single case study the atomic network is incredibly small and specific.

Facebook started not with the US or even a state, but with Harvard undergraduate students. Slack launched with a couple of dozen <100 person Bay Area startups. Ubers early focus wasn’t the US, or even San Francisco, but ephemeral moments like ‘5pm at the Caltrain station on 5th and King’.

Network effects begin in small, highly targeted networks where users can be captured and value can be created quickly.

But, users aren’t all equal, there is one particular type of user that is particularly important..

The hard side

Every network has a user that provides disproportionate value. Wikipedia has power editors, Uber has highly committed drivers, Youtube has content creators. eBay has people actually selling things. These are the users that make a network valuable.

As a rule of thumb with content apps (think TikTok) the 1/10/100 rule applies:

  • 1% of users create content
  • 10% actively engage with content (sharing, commenting etc)
  • The rest ‘lurk’ and get the benefits of these folks

Attracting that small number of highly engaged users that create network value is called the hard side. These numbers aren’t always static, ~20% of Uber’s drivers provide most of the rides, conversely just 0.02% of Wikipedia viewers contribute significantly.

Typically the hard side of a network is the supply side. On a marketplace that is sellers, on Uber that is drivers, on social content apps that is creators, on Airbnb it is hosts.

Each atomic network has it’s own dynamics for tipping points. The scale of these networks varies but it’s easiest to think of as a city for products like Uber, Facebook, Tinder or Airbnb. At what point does a city become a useful network? These can quantified and become a key metric for product launches.

The Tipping Point

Building a single, atomic network is the first step towards a large and dominant network effect company. Creating a single atomic network is a challenge, creating a repeatable approach to building these atomic networks in different circles, cities or groups is where things become really interesting.

How do you create this repeatable way of building atomic networks in new groups? There are three approaches described:


Many very well known products launch as invite only, LinkedIn and Gmail both began in this way. The advantages of invite-only give people who otherwise may not join a product a reason to join. Much like the friendship paradox your contacts are better connected than you are meaning those who initially join an invite-only network are likely to connect other well connected people.

Come for the tool, stay for the network

Many networks begin with a useful tool. Users initially join to get access to this tool and then use networked features as a secondary benefit. Over time the benefits flip and the network becomes the source of value.

Instagram is a great example here, the initial reason to download was a way to easily add filters to your photos. The sharing, profiles and social features were secondary. Over time this has flipped.

Paying for launch

In a more than a century-old tactic, paying for launch can incentivise both users and hard side to adopt. From evaporated milk companies handing out coupons in magazines (and using the incoming rush of sales to onboard grocers). To streaming content platforms like Twitch and Netflix paying content creators upfront.

Paying one, or both, sides of a network is a proven, yet expensive, strategy of moving through the cold start problem.


Named after Fred running his bare feet below his stone aged car to get around Flinstoning is exactly what you are imaging: making it seem like a network is more active than it really is to encourage users to join and gain value.

From theReddit founders in the early days posting content themselves (and then with bots), to PayPal’s early bots buying and selling on eBay and insisting on PayPal as the payment method. In each of these examples the company used their own effort or tools to make the network appear more utilised and valuable than it was to encourage adoption and early users to stick around.

Escape Velocity

Ok, so now you are huge. Millions of users and the network is creating a lot of value for both your users and hard side. Now what?

There are three complimentary forces that can begin to be exploited to maximise the value the network creates:

The engagement effect

Most users churn quickly. 70–90% of users becoming inactive immediately is normal. However, as a network becomes larger and more frequented by users it is something which can reactivate these churned users, balancing out the churn.

To exploit the network effect divide users in cohorts and use A/B testing to show causality, and strengthen the engagement effect. Rinse and repeat.

The acquisition effect

Product-driven viral growth is a magical condition where new users of your product drive other new users of your product. PayPal grew quickly because sending money directly to someone required them to sign up to receive it and then made it immediately easy for them to send money and so on and so on.

This effect is driven by these new users immediately becoming engaged with the product. So acquisition effect is driven by viral product growth and new user engagement through viral growth.

The economic effect

As a hugely broad generalisation, larger networks are more efficient than smaller ones. In the case of Uber, more riders can suppport more drivers which reduces wait times and increases utilisation, reducing network cost.

In software, companies with a product-led growth strategy attract users who are more likely to share the product and then benefit from upgrading. By making the upgradable features useful for the average user that increases the odds of an upgrade happening.

The Ceiling

Eventually the growth party stops. Maybe through competition, saturation or other forces. This isn’t abrupt, often their will be periods of fitful growth and contraction but things begin to level out. What do you do then?


A good problem to have: you are so popular you just saturated the market and you are not growing exponentially anymore. What can you do then?

Different companies take different approaches. Instagram focused on adjacent users, those who know of the product but aren’t highly engaged. eBay focused on new geographies to kickstart growth. Others acquire other companies with strong growth potential to expand the market opportunity they are playing in (as eBay did with PayPal).

Reduced marketing efficiency

Every marketing channel degrades over time and it more becomes more expensive to acquire users. Every product should continue to add more channels to their marketing distribution as they scale. But, network effects can be used to acquire users as well. Twitch

Reaching scale through paid acquisition is nearly impossible. Harnessing network effects in marketing is much better.

The Moat

You’ve weathered everything to here, now you finally get to reap the rewards.

A networked moat is hard to challenge directly precisely because it is so hard to cultivate and so much of the value is already tied up in the connections and familiarity with what exists there.