Launching your first atomic networks to solve the Cold Start Problem

As Chen explains in his book, one group of people in a network is often harder to acquire. Yet, it is still very important to do so because it will generate most of the business and user value. Chen refers to these distinct groups as the “hard side” and the “easy side.” Knowing which is which in your product is paramount to the success of your business.

Examples of hard and easy side

In two-sided marketplaces, the hard side is often the supply. On Wikipedia, the hard side is the contributors — who dedicate countless hours to creating content. The easy side is the readers. For Uber, the hard side is the drivers. Riders make up the easy side. In Slack, the hard side includes admins, managers, and group leaders who manage channels and team members. The easy side is the employees who communicate.

Characteristics of the hard side

As a product builder, it's essential to understand the underlying reasons that make the hard side challenging to attract. By uncovering these causes, you can better distinguish between the hard and easy sides and develop more effective strategies for acquisition, engagement, and revenue.

• Asset Scarcity: The asset of an online network is what is being exchanged. On Uber, it’s the ride. For Wikipedia, it’s the article. Assets can be more or less rare. Drivers are not always available to drive at odd hours. And only a few people are willing to write for free on niche topics.
• Value Expectations: The hard side may have high expectations about specific product features, number of people already on the network, or revenue. Understanding what are the expectations of your hard side is crucial.
• Effort: Since the hard side invests more effort, attracting these users is naturally more challenging.

Each side impacts the business differently

Knowing which side is which in your product is paramount. Each side will affect your business differently. Acquisition strategies must be tailored to each side. A two-sided marketplace for cars like Clutch has distinct acquisition channels and marketing messages for buyers and sellers. Engagement metrics and levels also vary by side. For example, if you are measuring time on site, the hard side will often spend much longer online. Revenue, which often comes from an interaction between the hard and the easy side, is also disproportionately impacted by the hard side’s activity. Much more revenue comes from the activity of a YouTube creator versus a viewer.

In an asymmetric network, even if the network is healthy, one side will be larger than the other. For example, Wikipedia is quite asymmetrical: only about 0.02% of users create content, while the rest are readers. Assessing the asymmetry of atomic networks is important because it informs how you build and manage your network.

What causes network asymmetry?

In two-sided marketplaces, network asymmetry can arise from the need for choice. An example is Airbnb. Overtime, the growth team learned that when they open to a new location, they need a minimum of 300 listings. Another cause of network asymmetry in two-sided marketplaces is asset scarcity. Consider an online auction where artists sell unique pieces to their fans. Artists are the hard side. Because what they sell is so unique, the atomic  network is just one artist and its many fans.

Social networks also tend to exhibit asymmetry. Here, the difference stems from the effort needed to create content. For example, a TikTok user has to learn dances, arrange lighting, and record several times. Compare that to Snapchat where users take and send ephemeral pictures.

How network asymmetry impacts a business

Asymmetry influences growth and network monitoring. Kickstarting an asymmetric network will be challenging, especially if the larger side is the hard side. You may also need to monitor very different indicators for each side of the network. This extra effort required for managing asymmetric networks has a unique advantage though: defensibility.

Estimate asymmetry for your atomic networks

You may already have a sense of how symmetrical your own network is. Here are some additional methods:

• Assess the need for choice. If your network is a marketplace, do users benefit from a large selection?
• Evaluate asset scarcity by examining how unique and hard-to-replace the assets on your network are.
• Analyze how you track engagement. Are you tracking widely different metrics for each side of your network?
• Look at revenue concentration. Are there users responsible for 10x, 100x the revenue compared to other users?

It is not an arbitrary number

Defining the minimum size of your atomic network will guide your execution and planning. The smallest atomic network is the smallest group that supports self-sustaining interactions, where users get real value from the network.

To estimate this right size, start with user research. Interview the easy side to learn how much content or how many listings they’d like to see. Interview the hard side to understand their expectations — how quickly and how much they hope to sell, how many viewers they anticipate. Competitive analysis provides additional guidance. What are your direct and indirect competitors doing? How many listings per category or videos per topic do they maintain? Look for patterns that indicate their minimum viable network.

The asset impacts network size

The asset is what users exchange on the network. In a marketplace, it’s the item; on a social network, it’s content—tweets, videos, or DMs. Focusing on a specific assets will make your atomic network smaller.

Consider these real-life examples. In an online auction house, kickstarting the network is easier when it focuses on a specific type of collectible. On Slack, the asset is messages—all centered on work and even a single team or project. This specificity allowed Slack’s atomic networks to be small, needing only three people to create a viable team, according to Slack’s founder. When Airbnb launched, it famously targeted tech conference attendees, narrowing the asset along three lines: purpose, location, and time.

Features also impact network size

Simplifying your product can reduce the number of users needed on each side. For example, if you’re building Slack and launching with DMs, channels, a directory, and apps, your atomic network will need people to activate each feature. All users will chat, but only some will manage the directory or build apps. By starting with fewer features, you’ll reduce the initial user count required to establish a viable network.

The characteristics of a healthy atomic network

A network consists of two main components: the product and its people. A healthy network has six defining characteristics that map back to the two components. It includes active users, experiences organic growth, and avoids decay, it generates revenue, its assets are regularly exchanged, it minimizes user “zeros,” or negative experiences, as Chen describes them.

What to track

Because a network has 2 sides and the 6 health characteristics, there are 12 key metrics to monitor. For each side, you want the following KPIs:

1. Size: The number of active users on this side of the atomic network.
2. Acquisition: A measure of the conversion rate such as from homepage visit to active user signup.
3. Churn: The percentage of active users who became inactive.
4. Positive engagement: Chen calls this a Magic Moment. Example: a video view for the easy side or a video submission for the hard side.
5. Negative engagement: Chen calls this a Zero. This could be searching for a topic and not finding it, or the number of unanswered messages, or unsold items.
6. Revenue: Revenue per user (ARPU). This could stem from video views and ads, or sales.

The line in the sand

The question is also how much. Even setting rough targets is beneficial for a few reasons.

Network size targets will give you your top line goal. By setting a target conversion rate, you can estimate the time and budget required to reach your minimum network size. Estimating a churn rate can help budget ongoing growth efforts. Don’t disregard the impact of churn and expect high rates as new users will often experience “zeros” while the network is sub-scale.

The early days of an atomic network are crucial. It feels like a acrobat, trying to maintain 4 stacks of plates while driving a bicycle. There are two sides to manage, different sets of needs and growth strategies. New people arrive, and some leave. Let’s see how to manage this initial phase.

Stabilize the hard side

As we talked about earlier, one “unit” of the hard side drives the engagement of a large portion of the easy side. Consider a content creator with 5,000 followers who suddenly stops posting. How will that impact their followers’ engagement? This simple example highlights an important network dynamic: the growth, engagement, and retention of the hard side has an outsized effect on your network. So in the early days, a healthy obsession with gathering and stabilizing it, is not a waste of time.

Making the hard side happy might require extra features, and that’s okay. Tinder is a great example of this approach, with numerous features designed to retain the hard side. Connect with Facebook increased trust by showing friends in common. Built-in messaging enabled the hard side to communicate without sharing phone numbers. Tinder also used swiping to learn users’ preferences progressively—without the need to fill out long questionnaires.

You might need to steal the hard side away from other networks. For example, Uber focused on unlicensed drivers during the second phase of their growth. Since these drivers were underpaid, Uber quickly attracted this underserved group. Like Uber, start by focusing on a subcategory or an underutilized asset from other networks. This focus will make your value propositions more compelling and your marketing easier.

Coordination and Speed

Unfortunately, focusing on one side one month and switching to the other side the next won’t work. You need both sides in simultaneously. For marketplaces where choice is needed, the kickstart might look a little different. As Chen says, "Supply, Demand, Supply, Supply, Supply."  First, reach critical mass on the hard side and then work to attract the rest of the network.

Speed is essential. You need to reach critical mass on each side as fast as possible. Why? Because until then, most users will experience too many "zeros" and not enough "magic moments," causing your network to keep self-destructing.

Pricing strategies that foster network effects

“Networks love to be free,” says Chen. As a product builder, you want to generate both engagement and revenue. To accomplish this, your pricing model must allow users to get some initial value without payment. Initial is the keyword, here are the two main ways to do that.

The fee. In marketplaces where assets are transferred for a price, your initial business model should be to take a fee. This is an good model to start with because it does not block initial engagement and revenue is generated only when both sides are happy.

The throttle. Crafting the right business model for non-marketplace networks is more challenging. Here, let’s use another engine analogy. A good approach is to throttle your product, that is: define a usage cap that serves as a paywall. Zoom, for example, allows you to get 40 minutes free to showcase its capabilities. Slack lets users exchange up to 2,000 messages for free, with searchable access—until they hit the limit and need to pay. Determining the right threshold can be tricky and may require some initial guessing. But as soon as possible, learn how much usage leads to long term user stickiness, and place your throttle level right after that level. After!

The cold start problem never completely disappears. Throughout your product’s life, continuous investment in tools for monitoring and rebalancing is essential.

The forces that influence network health

Atomic networks are either self-sustaining or self-destructing. Why?

• People. The more people on the network, the better the engagement because magic moments happen more frequently. Engaged users lead to more word of mouth, which improves growth and brings even more people. The opposite is also true: “zeros” lead to inactivity, which drives churn and network decay. Changes in user characteristics also impact the network. People are naturally diverse, and for niche networks, there is simply not an infinite number of people with the same interests and behaviors.
• Product. Inevitably, you will introduce new features that enhance your product. These great features will modify how users engage with the network, progressively impacting acquisition, engagement, and revenue.
• Assets. A network is made of assets being transferred —such as messages, videos, or physical items. Eventually, you may face asset-related toxicity, such as bad content or fraudulent listings.
• Outside world. External events also impact network activity. For example, conferences in town increase demand on Airbnb. Weekends see lower activity on Slack. Seasonality affects platforms like Europe’s largest boat rental marketplace, Click and Boat.

How to rebalance your network accordingly

These forces constantly affect your network, causing one side to fluctuate in size and activity. For example, on a content network, you may notice that a sub-topic is constantly underserved. In a marketplace focused on art, you might find that many users are trying to sell antique chairs, but buyers are scarce. 

As a result, you will regularly rebalance your network. One way to do that is with product design, For example, you could throttle or facilitate access to a specific side or asset category. Subsidies are also a good way to rebalance, as seen with Uber’s surge pricing strategy. Growth is also an obvious tool here; targeting and acquiring specific network subcategories can help with rebalancing.

Zeros are your first leading indicators

Monitoring “zeros” and churn is critical for spotting issues. Zeros are an early sign of decay, allowing proactive intervention. Initially, data may be limited, so you may not know any other leading indicators for decay or growth—and that’s okay. Focus on Zeros and being able to respond to them, network decay and other lagging indicators quickly.