The Cold Start Problem

The Cold Start Problem

I was initially skeptical reading Andrew Chen's Cold Start Problem. I hadn't followed him closely, and my first impression of Andrew was that he was a growth hacker who didn't really focus on the finer points of product at all.

Boy was I wrong. This book is an excellent read. If you are curious to analyze the way viral products are built and spread, this is the book for you. It spans not only consumer apps, but ties ideas through to the enterprise as well.

Network effects

Andrew starts off with a discussion of network effects.

Everyone is always talking about how network effects will cause their particular product to win... but what really is a network effect?

There's a difference between the network and the effect.

A network are the people involved with a product. The network effects are the parts of the product that connect people, to do commerce, talk, communicate, or otherwise interact.

If you don't have a connection, or if one person can get a lot out of the product without another... then you don't have a network effect!

Andrew hopes that The Cold Start Problem is the definitive book on network effects, which he points out aren't studied nearly well enough.

Meerkat's Law

Most of the time, people like to model networks via Metcalfe's Law, where the value of the network increases exponentially with the number of nodes.

There's something intuitively nice about this explanation, it makes sense that the more people you can reach in a network, the more value you get.

But that doesn't really model what's going on! Does the billionth person to join Facebook really add as much value as the millionth? Probably not.

Instead, we should borrow our idea of networks from looking at nature.

via Andrew Chen,

The network 'take-off' happens at the Alee Threshold, a point where the network can become self-sustaining. This is different for different types of networks. For Airbnb, it has to be much higher (a city) than for Zoom (two people)!

What's interesting is that there are a number of events that can cause the network to grow or to peak. Much like a population's carrying capacity, networks can saturate the resources they have available.

Atomic Networks

The type of tool you're starting will require a different type of 'atomic network'.

For slack or zoom, you only need a network of 3 or 2 people respectively. For Airbnb, you need 100 listings in a city. For Uber, you'd need 300 drivers.

When Bank of America launched its first credit card, it started with Fresno. Why? It had 45% markershare, and Fresno was small enough that they could just Mail everyone a credit card. They could build demand and then go to merchants.

Uber and Lyft need this too, they originally targeted times "drivers at 5pm at the 4th and king Caltrain".

I hadn't really considered this idea before, but now that I think about it, I see it everywhere. Product builders need to carefully consider what their atomic networks should be.

Start with the hard side

One part of your network will be harder than the other and will require more time.

  • producing content vs creating (tiktok, facebook, twitter)
  • driving vs riding (uber, lyft)
  • building games vs buying them (steam)
  • management vs users (slack, zoom)

Try to always start with the hard side of the network and understand what appeals to them. Generally this is utility, status, or something else entirely.

For most networks, the areas to focus are...

  1. supply
  2. demand
  3. supply
  4. supply
  5. supply (forever more)

Do one thing well

Network products will generally do one thing really well. In Zoom's case, that's way better video quality. In Slack's case, that's providing channels.

If you find yourself adding feature after feature, it's a sign that you're probably missing some big important thing that you might care about (a la Kevin Kwok's Atomic Concepts).

Anti-network effects

Andrew recommends looking at areas where users hit a 'zero'. Some interaction where they aren't getting any benefit of the network.

A good example for Uber is when a user logs on and doesn't see any cars nearby. If a user hits too many zeroes, then they won't log on. This reminds me of Amazon not having products in-stock on their product pages in Working Backwards.

Come for the tool, stay for the network

This phrase comes from a famous Chris Dixon essay, and it's a good reminder

  • Figma - you can do stuff on your own, but it's better with colleagues
  • Dropbox
  • Gmail
  • Snapchat/Instagram

Power of Network Effects

There's a good framework for leveraging the power of an existing network. Instead of thinking "I have a lot of users, I should be winning", it's worth thinking about how you can really leverage those existing users to grow your business.

Andrew presents three ways to do this...

  • Engagement: you can re-engage existing users if stuff is happening on the network that they care about. It's way better to message users about something that is relevant to their interests vs a generic "come back" marketing message.
  • Acquisition: users can invite their friends if you offer various referral bonuses. It typically will only work if the product is good enough, but it can help move the viral growth factor reliably.
  • Economics: there has to be some aspect of connecting users on a network generates value for the network. Maybe that's transaction fees, viewing content, or paid seats. Whatever it is, the mechanisms and number can be adjusted to generate more growth.

Case studies

Tinder – apparently Tinder found the way around the cold start program by hosting parties at USC. For the very first party, you had to download the app to get in, they had a bouncer at the front door who checked for everyone. The next day, everyone who attended had an 'atomic network' of the people they wanted to meet at the party.

Gmail – gmail was originally invite-only because their servers couldn't handle more users! It turned out to be a good strategy to build hype. I'm generally a little dim on the view here on invite-only, but it definitely built some exclusivity with both gmail and clubhouse.


Onboarding as dinner party – there's an interesting analogy that new users coming to an app are sort of like people arriving at a dinner party. They are much more likely to feel at home if they are welcomed by someone they know, and if the guests are engaging and interesting. Andrew argues that new apps should think along similar lines when crafting their initial network.

Networks copy themselves – if you have a network that requires inviting people, the original cohort will tend to invite other people like themselves who are excited about the idea. It's an interesting idea that curation leads to replication.

Big-bang launches don't usually work – too many founders try and do a giant big-bang Apple-style launch. They believe this is the way (along with bundling) to win the market. I've been guilty of this myself. It's important to remember that it almost never works. Instead, users refer other users.