Great book if you are building a network platform
Charles Petrini-Poli
Healthcare Leader | Growth Strategist | Sales & Operations Executive | Consultant in Healthcare Innovation
I know many of my connections on LinkedIn have software projects in development. I forward this book summary as the concepts are valuable.
This book, The Cold Start Problem, is the best business book I've had the pleasure of reading in some time. The earlier chapters, focused on building the first atomic networks, were the most valuable and informative — likely because of my position as an early-stage startup founder. However, the later chapters may be of more interest to you if you're a leader in an established platform company, dealing with competition, or stalling growth.
The Cold Start Problem | By Andrew Chen?
In The Cold Start Problem, Andrew Chen draws on his experience as "Head of Rider Growth at Uber" to analyze network effects. The book answers the top-of-mind question for leaders at platform companies: how do you use network effects to scale your product?
Chapter 1: Network Effects: A Primer
A network effect describes what happens when a product gets more valuable as more people use it. Conversely, a networked product without a network is a useless thing — a telephone with no one to call, a ride-sharing app with no drivers, a network of rooms to rent with no renters.
It's harder now than ever to launch a product — the battle for attention is fierce, and the biggest winners crowd out the rest. This highlights the importance of network effects — you can copy all of Instagram's features, and you can leverage the same marketing channels, but you can't easily convince hundreds of millions of users to switch to your product.
Networked products started as a B2C phenomenon, but increasingly there has been "consumerization" of B2B apps — with apps like Slack and Notion getting their start by gaining a few ardent users within a company then spreading to the entire organization.
Meerkat's Law
The dot com boom (bubble) brought theories that have outlived their accuracy. Chief among those was Metcalfe's Law, which states that the value of a network is proportional to the square of its participants.
Metcalfe formulated his law based on his experience of selling Ethernet. However, Chen argues that it is not clear that a network like eBay or Paypal is equivalent to "compatible communicating devices" like Ethernet-enabled devices.
It does not consider how overcrowding leads to degraded experiences, nor does it consider how different network participants (typically sides of a market) bring variable value to the network. Additionally, perhaps most importantly, it does not offer practical advice for starting and growing a network.
Chen puts forward an alternative — Meerkat's Law. Meerkats benefit from living together for safety and resource gathering like other hypersocial animals. The value of being a Meerkat increases as more Meerkats join the gang. However, overpopulation negates these effects causing the population to plateau or collapse. Chen borrows terms from ecology to put forward his theory of network effects.
Allee Threshold
The point at which animals would be safer and ultimately grow faster as a population. This is the ecological version of a network's "Tipping Point."
There is a circular dynamic above and below the Allee threshold where the analogy with technology is clear — not enough people on a messaging app leads to people deleting it. This means fewer people on the app, which prompts more deletions. The inverse is also true.
Carrying Capacity
A limit to growth — beyond which resources become exhausted (servers slow down, the choice becomes too much, etc.). In ecology, this is called the carrying capacity of the environment.
Cold Start Theory
Chen calls his framework Cold Start Theory — his core motivation is to describe what happens as a network forms and evolves and offer practical, actionable advice to propel a networked product from one stage to another. There are five stages of network growth:
1 . The Cold Start Problem
Most new networks fail; therefore, network effects hurt most startups. Chen calls these "anti-network effects." The basics of the cold start problem are getting the right people and content on the platform simultaneously.
2. Tipping Point
It's not enough to have a single network. Every network is a network of smaller networks. As each new network comes online and grows, building and scaling the next one becomes more straightforward. As a result, the networks tip faster and faster; they are dominoes with overwhelming momentum.
3. Escape Velocity
Chen argues that classical theories about network effects are wrong — it's not one special effect where once you have a network and it's growing, you're set for takeoff. Instead, it is three distinct underlying forces:
a)?Acquisition Effect?lets products tap into the network to drive low-cost, highly efficient user acquisition via viral growth.
b)?Engagement Effect?— which increases the average engagement of network participants as the network grows.
c)?Economic Effect?— which improves monetization levels and conversion rates as the network grows.
4. Hitting the Ceiling
The classical theory says that when a product has strong network effects, it's happy ever after. However, employees will tell you that the network is trying to grow and rip itself apart. This is where the network has "hit the ceiling" and growth stalls. Networks hit many ceilings, which are resolved, causing growth spurts until the next ceiling.
5. The Moat
The final stage involves using network effects to stave off competition. However, all competitors can, in theory, create their networks to compete — this is "Network-Based Competition." Often it's David vs. Goliath, where the smaller network must pick off niches within a more extensive network and build atomic networks that are highly defensible. The incumbent uses its larger size to drive high monetization and value for its top users and fast-following niches that seem to be proliferating.
The Cold Start Problem
It's easy to imagine networked products as overnight success stories, but they are usually not. Each must start with just a single network and grapple with "anti-network effects." These are the opposing forces that drive new networks to zero.
The Antidote to the Cold Start Problem
It is crucial to have the right users on the network — e.g., ten people using Slack all from the same team is better than ten random people in a large company. Therefore, density and interconnectedness are critical.
The Atomic Network
The anatomic network is the smallest network that can stand on its own. It needs to have enough connection density and stability to break through early anti-network effects and ultimately grow on its own. So do things that don't scale, focus on building density, ignore the objection of market size, do whatever it takes — even if it's unscalable or unprofitable in the short term.
The Hard Side
An essential dynamic at play in every network that only increases over time: a minority of users create disproportionate value and wield excessive power. This is the hard side of your network — they do more work and contribute more to your network, but they're harder to acquire and retain.
Even for networks where it's not clear that there are two sides, there is still a challenging side. e.g., messaging networks — the hard side are the active, extroverted users who initiate conversations and organize get-togethers.
Because the hard side is so critical, it's imperative to hypothesize how the product will cater to those users from day one. How do you find a problem for the hard side of a network where they will be engaged, but their needs are unaddressed? The answer is to find hobbies — what are people doing on their nights and weekends? This represents?underutilized?time and energy. Uber's supply side is underutilized drivers and cars; Ebay's is underutilized objects.
Avoid Zeros
When a product has solved the Cold Start Problem, it becomes apparent that the experience feels like magic. A car, out of nowhere, arrives to pick you up in 2 minutes; the exact product you want is the first item for your search.
To gauge magic moments, you must first start with the opposite of magic, the moments where the network has broken down. Uber called these moments "zeros."?A zero at Uber was the worst experience you could have when a rider opens the Uber app with the intent to book a ride — but there aren't any drivers in the area.
Every product has zeros — for a workplace collaboration tool, it might be stale or missing documentation. The coworker you need to reach doesn't have or hasn't activated their account for a workplace chat tool.
The actual cost of zero isn't just at the moment where it's experienced; it's the lingering destructive effects afterward. Users who zero often churn, and worse, they believe the service isn't reliable. Until this lethal force is dealt with, the network cannot get off the ground. To consistently ensure that people never experience zeros, the network needs to be built out substantially and active.
The Tipping Point
Beyond building a single network, you have to scale from 1 to 2 and then many more. The broader network effects that include viral growth, increased stickiness and strong monetization — start to kick in only once you scale. Once this becomes repeatable, the network will hit the tipping point, the moment a product can grow to take over the whole market.
Chen puts forward four strategies to scale out from your initial atomic network.?
Invite Only
The invite-only tactic is often described as taking advantage of the fear of missing out — FOMO — but that's not the core driver. Instead, invite-only mechanics provide a better "welcome experience" for new users, like walking into a party where you already know someone there. Mathematically it works because the most connected users are invited earlier, and in turn, they tend to invite other highly connected people.
For networked products, the density of the network — who is on it, why they are there, and how they interact with each other — is as important as its product design. Starting with a deliberate point of view on who is best for your network will define its magnetism, culture, and ultimate trajectory.
Come for the Features, Remain for the Network
It is best to initially attract users with a single-player tool and then, over time, get them to participate in the network. The tool helps to get the initial critical mass. The network creates long-term value for users and defensibility for the company. The tool can be used to "prop up" the value of the network effects curve when the network is small.
Pivoting users from a tool to a network can be challenging. Sometimes only a tiny percentage will make the transition because it requires them to change their behavior — to click on a notification or a piece of user interface introducing them to the network — and then they have to stick. Many people can get stuck on just the tool. Not every feature can be a social network.
Paying for Launch
When facing a chicken and egg problem, buy the chicken.
领英推荐
Paying the hard side is risky and should only be executed right. While establishing an initial network usually doesn't make sense for an under-resourced startup to throw much money to get started. Instead, teams often focus on basics like figuring out the right market before reaching the financial lever. When you can launch a product's initial atomic network, economic levers can rapidly accelerate the market's speed hitting the tipping point.
Flintstone
Like Fred Flintstone using his legs to power his prehistoric car by running, you manually keep the network alive and healthy. Use automation to help.
Flintstone props up the network with highly manual work while "Come for the Tool, Stay for the Network" props it up with software. Like with the tools approach, it's essential to have an exit strategy — you must have a plan to switch to a fully automated process.
Escape Velocity
When your product succeeds and starts to scale, it's often called hitting "Escape Velocity." The mythology is that the product begins to hockey stick, going up into the right forever. But it's not so simple — in reality, the journey isn't over; instead, the focus changes. In this stage, the challenge quickly becomes maintaining a fast growth rate and amplifying a successful product's network effects. To explain, Chen breaks down network effects into three distinct forces.
Acquisition Effect
Viral products attract people to the product, amplifying the network's reach. Viral growth is the?land and expands?strategy — because it attracts new atomic networks to the platform. For example, the engineering team may all be using Confluence, but they send a document link when they need to loop in the Marketing team. If Marketing like the product, they may consider adoption.
Engagement Effect
Because of the engagement effect, networked products have some of the best retention curves. The network keeps users engaged for longer, rather than any product feature. As a result, what starts as infrequent usage can transform into a daily habit with the right nudges.
The Engagement Effect can reactivate "dark nodes" (deactivated users) in your network. When users first signed up, they may have had a poor experience and were deactivated. The Engagement Effect is helpful to reactivate for two reasons:
"Jack just signed up for…" or "Jill edited super important document" are effective marketing messages for reactivation.
Economic Effect
The economic effect is partially the data network effect, but it encompasses other elements.
The cost per unit is often remarkably high if you subsidize the hard side to encourage network growth. On the other hand, a more extensive network can decrease the cost. E.g., a guaranteed hourly rate for Uber drivers might be recovered if they can make enough trips per hour. But if they can't find any riders, it's pure cost.
Moreover, as the network grows, you gather data to understand the right way to subside — Youtube doesn't pay flat rates; it understands more about the right financial incentives to reward creators who make engaging videos.
All networked products rely on conversion, and the key is to create features that benefit everyone and promote a rich network. The premium feature of fully searchable history for Slack is essential as the company purchases more licenses. Increasing the likelihood someone will pay for everyone to have it. For Tinder, the "Super Like" is valuable when the network is crowded to stand out. For marketplaces, the chance of someone wanting any particular product increases when the network is denser.
The Ceiling
The ceiling is where growth stalls for networked products — the reasons are numerous: saturation, spam, low intent users, and poor clickthrough performance.
Saturation
Networks reach a point where adding more of the same to a network is not adding meaningful growth. Everyone who could be on the network is on it, and diminishing returns are going from seeing 1000 to 2000 product listings for every search vs. 10 to 20.
The term often used is "market saturation" — this is a misnomer — it's more accurately described as "network saturation." The key to unlocking new growth is to stack lots of exponentially growing (but small) networks onto the leading network — this may be through pushing the demand side into new geographies or through extracting more (or new) value from the existing network.
New, Adjacent Users
These are layers to the cake — the next mini-network for whom the product is not quite there yet, but could be soon, provided the right people are in the network, the right features are in the product, etc.....…
New Formats
Allowing the same users in the network to interact in new ways — eBay's "Buy Now" feature unlocked growth because people wanted confidence in the price they'd pay and the fact they'd get the item. It's also more appropriate for a broader range of things — rare collectibles need an auction for price discovery, a new hardcover doesn't.
New Geographies
This works well when the markets are adjacencies, so there is an overlap in the network — e.g., launching a social network at Harvard then MIT makes sense because there is likely to be overlap in friendship groups. Be aware that similar networks may be more different than you think!
Context Collapse
If it starts with a focused atomic network, every network begins with a concept like "netiquette" — a shared context and culture that governs behavior.
Context collapse is what happens when too many networks collapse into one. The photos you shared of that party do not want your boss or parents to see. Likewise, you talked about the company when your team was using Slack is not something you want the CEO to see.
A Network of Networks … of Networks (or fighting context collapse)
Messaging apps like iMessage & Whatsapp are resilient to context collapse. Even Slack manages to mitigate it since conversations can shift to dedicated channels, tiny networks with their context. Other products have their versions — Facebook groups, "Close Friends" on Instagram.
However, this is not a silver bullet, it fragments the network too much, and discoverability and underutilization become a problem.
Downvotes
Networks that grow rapidly and uncontrollably are vulnerable to spam and trolls. The best defense is to use the network itself to fight this off. Community moderation backed by a framework codified in software — managed by the network operator can effectively prevent or minimize bad actors.
Overcrowding
For communication tools, you suddenly have too many messages to deal with. There is be too much content for social products to keep up with the people you care about. Or for marketplaces, too much choice makes you less able to decide.
Networks must deal with this to surface the right content to their users at the correct time. As a result, they will often start with manual curation, popularity rankings, and algorithmic curation.
Network operators need to be mindful of the "old money" effect — where prominent players get their content promoted by having a large following already. If the best content is not encouraged because the creator is small, it incentivizes new creators to go elsewhere.
The Moat
Traditional product moats are generally branding or a unique business model. Networked products have a different moat — how much time, effort, and capital does it take to replicate the product's features and network. Although feature replication is a tractable problem, network replication is often not.
Competitors are seeking to enter a market face an even harder cold start problem because the low hanging fruit has been snapped up. Yet this moat may be limited. For example, Uber owning the SF market doesn't mean you can't compete and win in the Boston market — network effects are hyper-local. In contrast, AirBnB's moat is much more robust — due to the nature of travel.
The Battle of Networks
Networked Products lean toward "winner-take-all" dynamics. The battle is not on features — it's on creating dense, highly engaged networks — contrast Facebook's launch with well-connected college students vs. Google+'s big bang launch (extensive network, loosely connected).
Network Collapse
The maturity of the market dictates the nature of competition. In the early days of a category, everyone can grow their network. Later on, competition becomes zero-sum.
Cold Start Theory predicts that competition creates a virtuous and vicious cycle —network effects boost the winner and generate adverse consequences for the losers.
Cherry Picking
Craigslist is the ultimate network of networks — many billion-dollar companies have been built just focusing on one part of what it offers. Some features of the network are less well served than others. This is the upstart's advantage — they can cherry-pick the one desirable use case that the incumbent poorly defends.
This is symbolic of the Innovator's Dilemma. The concept of atomic networks provides the clearest goalposts for an upstart network — it's all about splitting off a higher-density atomic network. Part of why cherry-picking is so dangerous for incumbents is that the upstart network can acquire an entire set of users who have been conveniently aggregated on their network.
Big Bang Failures
The "big bang" launch is often the strategy of more significant players who use their size to overwhelm smaller competitors quickly. Intuitively this feels like it should work; it's an asymmetric advantage. However, it's often ineffective for networked products because it creates many weak networks instead of a few strong ones.
Startups have an advantage -- they can begin and grow in a small market (one that wouldn't move the needle for a larger company) — this makes them dangerous competitors.
Bundling
Using an existing network to create the new atomic networks can be successful. For example, post-acquisition, Instagram used the Facebook social graph to serve their customers in this way better. However, Bundling isn't always so practical. If the product is no good, you get the same effect as a big bang launch — you may contact people to try your new product/feature, but they won't stick around.