andrewchen

Why you’ll need to rethink your user growth strategy
Downturns fundamentally rewrite the industry’s strategy (and expectations) for user growth. In a bull market, the focus is on top line growth. You often want 2-3x YoY for a a new product in its first few years, and even faster when its right out the gate. High growth and high burn are fine. Because if you need to spend a lot of money to get there, whether through paid marketing or partnerships, you do it… after all, you can just raise more money, right?

But in a bear market, the answer changes: No. It turns out, you won’t be able to just raise more money to keep going. No, you can’t just expect to hire dozens of engineers, regardless of progress — particularly when hiring freezes are coming into effect. For startups, the bar for raising the next round just went way up, as many investors are waiting out the turbulent market. This means the strategy for user growth just went from “as much as possible” to “efficient, profitable, productive” in just a few quarters.

What are some ways you should be rethinking your growth strategy? Here’s some things every team should be thinking about:

• Embrace the new normal
• Cut your marketing spend
• Laser focus on your engaged, high LTV users
• Live to fight another day

I’ll unpack some of these as we go.

The new normal
Efficient growth is now the key focus for product teams. During a bull market, the primary metric that people talk about is just top-line growth — what’s your year-over-year growth rate. Some of the truly eye-popping growth rates might exceed 10x YoY, often subsidized with investor money — as has been in the case with on-demand services.

However, the new normal is focused on efficient growth. Although there’s a floor for how fast a product has to grow to be interesting — probably something like 2.5x — there’s a much bigger emphasis on efficiency. What’s the best way to measure this? One metric that’s been recently popularized by David Sacks is the “Burn Multiple” — he defines it below:

Burn Multiple = Net Burn / Net New ARR

This puts the focus squarely on burn by evaluating it as a multiple of revenue growth. In other words, how much is the startup burning in order to generate each incremental dollar of ARR?

In other words, if you spend $10M and gain $5M more in annual recurring revenue, that’s a 2x burn multiple — which he grades as “Suspect.” The Burn Multiple metric is simple, but it’s precisely useful because it’s so simple. A lot of times, unit economics are hand-waved by product teams because some costs are excluded from the contribution margin or net revenue calculations that maybe shouldn’t be — like headquarters costs, real estate, and so on.

Burn multiple cuts through all that, since it’s just aggregate cash versus revenue, and it’s hard to hide anything with a metric so simple. And with this simple metric, it allows you also compare different companies, and potentially different scenarios for a given company, to figure out how to best reduce it.

To provide some benchmarks, my colleagues at a16z, Justin Kahl and David George, recently wrote an article on navigating the downturn where they collected some empirical data:

As you can see, the bar for what constitutes a good burn multiple goes up as revenue goes up. Naturally, you burn more upfront during the product development phase, and then get more efficient as the business gets scale.

From a product growth lens, the shift from topline growth to efficient growth means that you should be thinking about how much burn, how many engineers, and how much marketing is required to hit the milestones you want to reach. And the first questions to ask are often around marketing.

Cut your marketing spend
The first and simplest thing to do is to cut your marketing spend. And in a particular order:

• Keep the high ROI channels, cut the low ROI ones, even if they provide volume
• Focus on accountable spend, and reduce ones have a long/fluffy payback?
• Rethink brand marketing spend — do you really need it?

On the first point, every marketing/growth effort is built from layers of channels built on top of each other. The highest ROI tends to be channels like SEO, word of mouth, and other organic efforts. The next might be paid channels like newsletters, which are hard to scale but highly productive. Then there’s highly targeted paid marketing. Usually the lowest ROI tends to be broad targeting — particularly display ads — on large advertising networks.

Usually these layers are built over time, one by one, by growth marketing folks who keep investing and arbitraging 10:1 LTV/CAC ratios down to 3:1, then 1.5:1, before they slow down. There might also be ongoing marketing experiments to try new short-format video or otherwise. It’s time to unwind that. Usually each incremental channel might add more volume, but is rarely as efficient as the preceding efforts. There’s a diminishing returns — Law of Shitty Clickthroughs strikes again — as each layer is built. Instead, go back to the core.

The other vector to think about this is direct response versus brand marketing. Brand efforts are a great way to spend money without understanding its actual effectiveness to impact metrics, and it’s time to dial those down. Whether it’s large scale events, brand marketing, PR/comms, splashy videos, or otherwise — unless you can justify the costs, it’s time to reduce.

Either way, it’s time to retrench and focus on high intent, low CAC channels.

Laser focus on your engaged, high LTV users
In a hot market, there’s often a land grab to acquire as many users as possible. If there’s a goal to grow 10% in a time period, the pressure is often to grow 10% by acquiring a mass of new users — most of which will burn off from lower usage — when the better option might be to grow 10% by incentivizing higher engagement from existing, core users. At Uber, it was often noted that it was much faster to get drivers to spend 10% more time on the platform, so that there’d be more “supply hours” to react to demand — than to acquire 10% more drivers in a market. The latter would require a big marketing push, and might take weeks for the drivers to ramp up to the same level of engagement.

The reason why this dynamic exists — where the core users outperform new ones — is that there’s often a central segment of where the product is really working, and then an “Adjacent User” where it only kind of works. For example, early Instagram was working well with high-tech, urban users but not well at all within older demographics. Later, it wasn’t working well for Android users in emerging markets. But there’s often pressure to grow by capturing new segments of users, rather than improving on the core, which means that the users that come in through marketing channels are worse quality, lower intent, and less engaged than the core.

This can become a tradeoff between Marketing versus Product-Led Growth, where the former drives CAC, whereas the latter is built on product development costs. The advantage of growth driven within the product — whether that’s better user onboarding, high impact features, or otherwise — is that they impact a wide swath of users within the product. You can invest once and get benefits over a long period of time, and amortize costs across a large segment of users. I’d lean towards product, when possible, when the roadmap is clear on what to do. Obviously marketing spend and engineering time isn’t interchangeable, but reducing marketing budgets while maintaining/increasing product teams feels like a good trade.

Live to fight another day
The milestones required to unlock additional funding/headcount has almost certainly gone up, and specifically for startups, the bar needed to raise more venture capital money has gone up as well. Understanding these milestones will allow teams to fight another day, and coming up with a realistic plan is a key step.

The best way to understand how the bar has moved is can be shown by a chart in the aforementioned article shared by my a16z colleagues David and Justin, showing how the forward revenue multiple for public companies are down significantly. Meaning, you need much more revenue to justify the same valuation — what used to be a 15x multiple is now 7x, meaning valuations are down half even given the same revenue numbers.

Or said another way, when the valuation of public software companies gets halved, then the amount of revenue needed to justify the a valuation goes up double. (For early products that think more about Active Users or DAUs or otherwise, you can recreate these graphs based on $/DAU or otherwise — and yes, those are way down)

This is causing a domino effect in the industry. When you see a $2B public company cut down to $1B, then a $500M privately held startup is cut down to $250M, and so on. The tricky part is that for a public company, of course you have a real-time stock quote to see these valuation changes. For a tech startup, you raise new funding rounds every year or two. That means for much of the industry, the next round of a startup just became much, much harder, but we potentially won’t know for a year+ how much the bar has moved. Either way, this means fewer resources to hit the same hard growth goals.

The easiest way to flex to hit these elevated targets is to take more time, with higher efficiency. Teams have to buy more runway, focusing on better ROI and not a “high growth, high burn” mindset to hit the growth metrics. For startups who have recently raised, they’ll need to “catch up” on their most recent valuation, and additionally progress to justify the customary 2-3x jump in valuation between rounds. That’s the new bar.

Conclusion
The next few years are going to see a lot of change in the tech landscape, particularly for how teams think of growing their products. Much of the last decade has been focused on growth by any means — and investor subsidies, chasing volume via high CACs, have all played a key role. But in the next phase, efficiency means that we’ll need to retrench within the industry and talk about quality and efficiency.

There’s a myriad of complex trends intersecting at the same time: The new Apple privacy changes to ad networks, the potentially stingy venture capital landscape, the hiring freezes that are happening, how web3 plays out over the next few years, and so on. Just as product leaders had to reinvent their thinking to take advantage of the mobile boom, we’ll see them do the same in the coming years for the new environment that’s rapidly taking shape. In the meantime, it’s critical for teams to take a pause, figure out a new approach, and build towards the next boom.

PS. Get new updates/analysis on tech and startups

I write a high-quality, weekly newsletter covering what’s happening in Silicon Valley, focused on startups, marketing, and mobile.

Leave this field empty if you’re human:

Hi readers,

I’m so, so excited that my upcoming book is dropping in one week, on December 7!! I’ve been working on it for years, and am thrilled that it’ll finally be out.

The Cold Start Problem, as the book is called, is about the secret that drives many of tech’s most successful products. It’s the story of how messaging apps, marketplaces, workplace collaboration tools, multiplayer games, all share a common thread of being products that connect people with each other. And launching and scaling these products requires a mastery of “network effects,” one of the most-used but misunderstood jargon terms in the industry. My book aims to change that, systematically laying out concepts for startups and folks launching new products to consider.

One of the crucial concepts to understand this is the concept of an “Easy Side” and a “Hard Side” of a network.

To describe this more detail, below, I’ve included a full-length chapter called “Solve a Hard Problem” focusing on the idea that every network has an “easy side” and a “hard side.” Here’s how I define each side:

There are usually a minority of users that will create disproportionate value and as a result, they will have disproportionate power. This the “Hard Side” of your network. They do more work, contribute more to your network, but are that much harder to acquire and retain. For social networks, these are often the content creators that generate the media everyone consumes. For app stores, these are the developers that actually create the products. For workplace apps, these are the managers that author and create documents and projects, and who invite coworkers to participate. For marketplaces, these are usually the sellers and providers who spend their entire day attracting users with their products and services.

In order to make your product and its network launch successfully, it’s not enough to build a killer set of features. It’s also important to solve something crucial for the “hard side” of your network. Cheekily, I call this, “Solving a hard problem” — and I use the example of Tinder to illustrate this concept.

Hope you enjoy!

Regards,
Andrew

Chapter 8: Solve a Hard Problem (Tinder)
The hardest problem to solve in creating the first atomic network is, well, the hard side. Focus on attracting content creators to a new video platform, or sellers to a new marketplace, or the project managers inside a company to a new workplace app. The other side of the network will follow. The question is, how?

The answer is by building a product that solves an important need for the hard side of the network. Let’s look at online dating, which evolved over time to better solve the matchmaking problem that has bedeviled humanity since the beginning of time. Dating apps are network effects-driven products that grow city-by-city, and the more folks that join the network, the better the chances that people will find matches. But at the beginning of the product category, the experience was terrible, especially for the hard side of the network.

The problem of too many love letters
Online dating was invented at the beginning of the web, in the early 1990s. They were designed like newspaper classifieds, where men and women browse large databases of profiles, and could message each other if they were interested. Match.com and JDate were successful pioneers in this category, which worked despite its flaws. The classifieds-based design created a poor product experience since the popular members — particularly women — would become overwhelmed with a large number of messages, and they would struggle to reply. At a bar or club, potential suitors might be dissuaded if they saw a line of people waiting to talk to an attractive man or woman, but online, there was no such signal. So in turn, the experience for everyone else also ended up poor, because it seemed like no one would write them back.

The lesson is, unsurprisingly, that attractive people — particularly women — are the Hard Side of the online dating network. A few years later, the next generation of online dating would emerge, led by products like eHarmony and OKCupid. These products used quizzes and matching algorithms so that the system could decide who got which matches, and how often. This ensured women got fewer messages, and hopefully more of the right ones. And the men got more replies too, so that it didn’t feel like it was devolving into a copy-paste messaging exercise.

It wasn’t until 2012, at the beginning of the explosion in mobile apps, that yet another generation of dating apps would emerge. These apps, exemplified by Tinder, would innovate even further for the hard side of the network. I had a chat with Tinder’s co-founder, Sean Rad, about how Tinder innovated on the previous generation of products. He described the combination of new ideas:

The older dating sites made it feel like you were doing work, like you were inside the office. You’d go and do work emails during the day, then go home and write more messages at night. Only to prospective dates rather than work colleagues. Tinder was different — it made dating fun. You could sign up without filling in a bunch of forms. It’s visual, you just swipe back and forth, and you could take 5 minutes to do it while you were waiting in line or something like that. It’s a form of entertainment.[^1]

The other problem was how to wade through all the replies. In real life, you’re often introduced to potential romantic partners through friends, or you had a shared context — like work or school — that helped filter. For online dating, the most attractive members of a network needed some additional signals to help sort through their matches. Tinder did this by integrating with Facebook, and Sean also explained how the app was able to build trust:

Tinder started by making everyone connect their Facebook, so that we could show the number of mutual friends you had, which built trust. We also made it so that you could only be matched with people who lived around you — we used the GPS location from your phone, which was new. These were people with mutual friends living around you, the sort of person you might meet in real life! Connecting with Facebook also made sure you would never be shown to friends, or vice versa, if you were worried about that. This all created trust. Tinder also had built-in messaging so that you didn’t have to give out your number. If the conversation didn’t go anywhere, you could just unmatch without worrying about getting harassed.

And of course, the mechanic of swiping itself is a way to make sure people don’t feel overwhelmed. Whereas men tend to swipe right (that is, to indicate interest) on about half of women’s profiles — about 45% to be exact — the ladies in the product swipe on only 5% of profiles they see. As a result, women mostly match with the guys they select. However, if they feel like they are in too many conversations, they can stop swiping for a while and just focus on the messaging their existing matches. All of these insights made Tinder a much better experience for most important side of their network, solving one of the most important obstacles in the Cold Start Problem.

The Hard Side for marketplaces is usually the supply side
Marketplaces tend to revolve around its sellers. I’ve seen the difficulty of managing the hard side for rideshare first-hand, where drivers are the ones selling their time and effort in the market. For Uber, in any given market, so-called “Power Drivers” constitute 20% of the supply but create 60% of the trips. These are some of the most valuable users on the planet, as they are the core of Uber’s business.

Uber’s drivers are just one example of a broader set of workers that drive most marketplace companies. For marketplaces, the hard side is usually the “supply” side of the network, which refers to the workers and small businesses who provide the time, products, and effort and are trying to generate income on the platform. They use digital marketplaces as a side hustle, selling collectibles or coaching sessions, or otherwise. They do this often as an alternative to hourly jobs, of which there are nearly 80 million in the US. These are folks often living in the middle of the country, who work in hourly retail jobs that turn over 100% year over year, and are struggling for additional income. Marketplace startups often provide these opportunities to this group.

To solve the Cold Start Problem for marketplaces, often the first move — as it was for Uber — is to bring a critical mass of supply onto the marketplace. For a marketplace like eBay, you start with sellers of collectibles. For a marketplace like Airbnb, you might start with people with a few extra rooms in their place. For a social platform like YouTube, it might be video creators. For a more esoteric category, like Github, it’s helpful to bring on some prominent Open Source projects and key developers. But once the supply has arrived onto the network, it’s time to bring in demand — the buyers and users that will form the bulk of the network. Once that’s working though, it becomes all about supply again. Thus the order of operations, at least for most consumer-facing marketplaces, is “supply, demand, supply, supply, supply.” While supply might be easy to get onto the network early on through subsidies, eventually it will become the bottleneck. The Hard Side of a network is, by definition, hard to scale.

Uber had to get creative to unlock its Hard Side. Initially, Uber’s focus was on black car and limo services, which were licensed and relatively uncontroversial. However, a seismic shift soon occurred when rival app Sidecar innovated in recruiting unlicensed, normal people as drivers on their platform. This was called the “peer-to-peer” model that created millions of new rideshare drivers, and was quickly copied and popularized by Lyft and then Uber. Jahan Khanna, cofounder/CTO of Sidecar spoke of its origin:

It was obvious that letting anyone sign up to a driver would be a big deal. With more drivers, rides would get cheaper and the wait terms would get shorter. This came up in many brainstorms at Sidecar, but the question was always, what was the regulatory framework that allows this to operate? What were the prior examples that weren’t immediately shut down? After doing a ton of research, we came onto a model that had been active for years in San Francisco run by someone named Lynn Breedlove called Homobiles that answered our question.[^2]

It’s a surprising fact, but the earliest version of the rideshare idea came not from an investor-backed startup, but rather from a nonprofit called Homobiles, run by a prominent member of the LGBTQ community in the Bay Area named Lynn Breedlove. The service was aimed at protecting and serving the LGBTQ community while providing them transportation — to conferences, bars and entertainment, and also to get healthcare — while emphasizing safety and community.

Homobiles had built its own niche, and had figured out the basics: Breedlove had recruited, over time, 100 volunteer drivers, who would respond to text messages. Money would be exchanged, but in the form of donations, so that drivers could be compensated for their time. The company had operated for several years, starting in 2010 — several years before Uber X — and provided the template for what would become a $100 biillion+ gross revenue industry. Sidecar learned from Homobiles, implementing their offering nearly verbatim, albeit in digital form: Donations based, where the rider and driver would sit together in the front, like a friend giving you a ride. With that, the rideshare market was kicked off.

Nights and weekends
The key insight in the stories of Homobiles or Tinder is — how do you find a problem where the Hard Side a network is engaged, but their needs are unaddressed? The answer is to look at hobbies and side hustles.

There are millions of content creators, app developers, marketplace sellers, and part-time drivers that power the hard side of networks. They are smart, motivated, early adopters who are finding opportunities to make themselves useful. They are the developers behind the Open Source movement who have built Linux, WordPress, MySQL, and many of the other technologies that underpin the modern internet. They are the millions of eBay sellers that have created jobs and companies by buying and selling goods that people want. For photo sharing and messaging products like Instagram and YouTube, they stem from the countless amateur photographers and videographers that like to record travel, special occasions, architecture, beautiful people, and everything else.

What people are doing on their nights and weekends represents all the underutilized time and energy in the world that if put to good use, can become the basis of the hard side of an atomic network. Sometimes the army is built on people with excess time, but sometimes it is built on people with underutilized assets as well. Rideshare networks, for example, fundamentally depend on the underutilization of cars, which generally sit idle most of the time besides the daily commute and the occasional errand. Airbnb is built on the underutilization of guest bedrooms, second homes, combined with the time and effort of the hosts. Craigslist and eBay are built on letting people sell their “junk” – the stuff that people don’t value anymore – to new owners who might value them more.

Usually the Hard Side will continue to use Airbnb or TikTok because that’s where the demand is, and thus, are locked into the positive network effects on those platforms. However, the trick is to look closer — it is better to segment the Hard Side of the network and figure out who is being underserved. Sometimes this is a niche, like a passionate sub-community of content creators for makeup or unboxing that might be better served with additional commerce features. It could be a low-production quality, amateur part of the community, like those who are doing #whateverchallenge of the week, who would benefit from basic video editing tools. For networks that are derived from underutilized assets, it might be the niche of who like having new side-hustles every weekend to make money online. Or perhaps there is a new platform shift coming soon that feels niche, but might upend the entire ecosystem.

The idea is to start with these underserved segments — which may not be very attractive customers on their own, and to apply Clayton Christensen’s disruption theory. New products often disrupt markets by starting on the low-end of the market, providing “good enough” functionality, and growing them there. They use a different technology foundation that allows them to eventually roll up the market from low-end into the medium, and eventually into the core market of the incumbents. Or, there has been a recent trend in the opposite — products like Uber and email company Superhuman, where you start at the top of the market as a luxury product, and work your way down.

When we combine disruption theory with that of network effects, it makes even more sense – atomic networks often start at the low end in terms of functionality, in a niche market. But once they establish an atomic network, then often the Hard Side of the network is willing to extend their offerings and services to go into the next vertical — attracting an incrementally higher-end opposite side, which in turn, spurs it even further. Airbnb may have started with airbeds, but the same hosts that might be willing to rent out an airbed might be willing to rent out their room, or their entire apartment. This changes the potential nature of the supply in the marketplace, attracting a higher end demand-side, which in turn attracts higher-end inventory. No wonder today, Airbnb hosts a wide variety of high-end offerings, from luxury penthouses to boutique hotel rooms. In that way, network effects can play a key role in disrupting new industries — creating the momentum for a low-end atomic network to slowly build out into higher-end offerings over time.

The Hard Side of dating apps
Let’s go back to online dating for a moment — when viewed as a networked products, the apps bring together two sides in a romantic context. In that way, Tinder, Bumble, Match, eHarmony, HotOrNot, and the line of dating apps reflects something that existed as a human behavior for eons. It’s long been a hobby of amateur matchmakers to introduce their single friends to each other, both demonstrating a deep need for this service as well as the skills needed to make it successful. In the the modern age we have digitized dating, using algorithms to match people, dating profiles so that thousands of profiles can be swiped through, and real-time messaging to make communication easier.

All of these improvements are great for any product, but most importantly, they help attract and maintain the most desirable members of a dating network — the Hard Side. The matchmaking algorithms need to find them equally attractive matches, and the profiles they browse through must help them decide between princes and frogs. The in-app messaging experience has to cater to their needs, with an option to get out of conversations quickly if needed. Without these types of features, desirable people will churn from the product, degrading the network and worsening the experience for everyone else. The best products have to solve a problem for the Hard Side of the network.

While dating apps — and really, all networked products — need to find a value proposition for the Hard Side of the network, what about all the other users? Well, it’s a high bar, but you need to nail the experience for the rest of the network. You need to build a “Killer Product” that sits at the heart of any network.

PS. Get new updates/analysis on tech and startups

I write a high-quality, weekly newsletter covering what’s happening in Silicon Valley, focused on startups, marketing, and mobile.

Leave this field empty if you’re human:

Hi readers,

In just a few weeks, I will be dropping new book, The Cold Start Problem!! If you haven’t gotten your copy — if you are in the US, here are the relevant links:

• Amazon · Bookshop (support your local bookstore!)
• Or if you are international, go to the coldstart.com website to find all the pages for Europe, Asia, and more.

Thanks to all of you for the tens of thousands of emails, tweets, and preorders in support of my work. If you’re reading about this book for the first time, you might be asking yourself — what is the Cold Start Problem about?

Let me set it up in a few bullets:

• Many of tech industry’s most valuable products — Slack, Zoom, Instagram, Twitch, YouTube, and others — are at their core products that connect people to each other. People are connected for commerce, communication, collaboration, and more
• I’ve come to believe this is the secret of much of Silicon Valley’s success. I saw it first-hand at Uber, which scaled to billions in revenue, and also within startups at Andreessen Horowitz, which has funded companies from Github, Coinbase, and Figma to Clubhouse and Airbnb
• These types of products benefit from “network effects” — that they become more useful as more users engage. This is why they grow to be so powerful, and valuable. But also why they are impossible to get off the ground, because people won’t use products where their friends/colleagues aren’t already engaged
• The Cold Start Problem is a book a collection of case studies, from Tinder, Twitch, credit cards, Dropbox, and others — about the lifecycle of these networked products. How to get them started, how to scale them, and what it’s like to compete
• These frameworks are targeted at teams building new products, first and foremost. But it’s also for people who work in the business of commerce, travel, publishing, and many other industries that are getting reinvented by tech.

I hope that’s a good teaser 🙂

In previous posts about the book, I’ve hinted at its contents. But over the next few weeks, I’ll be more substance and previews of the content.

Today, for the first time, I’m going to provide a preview of the opening chapter of The Cold Start Problem — which is partially about my experience at Uber and Andreessen Horowitz, but also about why I began writing in the first place.

I hope you enjoy it!

Thanks,
Andrew
writing from Venice, CA

The Cold Start Problem

Introduction
It was 2015 in December, and on a Friday evening, the office was buzzing. Amid the vast, monochromatic corridors of Uber’s San Francisco headquarters at 1455 Market Street – two football fields worth of gleaming LED lights, light woods, concrete, and steel – the office was still mostly occupied at 8PM. Some sat at their desks quietly typing email while others debated energetically with colleagues over videoconference. Others were drawing on whiteboards, hosting impromptu jam sessions to tackle the tricky operational problems facing who-knows-what. And a few pairs of employees were walking up and down the main flow in 1-on-1 meetings, some in intense discussion and others just catching up.

Everywhere you looked, there were reminders of the global scale of Uber’s business as well as the international heritage of the team driving it. Colorful flags from every country hung from the ceiling. Conference room screens hosted videoconferences with colleagues from faraway offices in Jakarta, San Paulo, and Dubai – sometimes simultaneously! Flat screen TVs were scattered throughout the floor showing metrics, broken down by mega-region, country, and city, so that teams could monitor progress. The global culture seeped all into the naming conventions for conference rooms: Near the entrance, the names started with Abu Dhabi and Amsterdam, and at the far other end of the floor, ended with Vienna, Washington, and Zurich.

At first glance Uber might just look like a simple app — after all, the premise was always to hit a button and get a ride. But underneath its deceptively basic user interface was a complex, global operation required to sustain the business. The app sat on a vast worldwide network of smaller networks, each one representing cities and countries. Each of these networks had to be started, scaled, and defended against competitors, at all hours of the day.

It was in my role at Uber that I really came to viscerally understand networks, supply and demand, network effects, and their immense power to shape the industry. As you might imagine, the Uber experience had its ups and downs – it was a rocketship and a rollercoaster, rolled into one. I’ve come to call it a “rocketcoaster” experience, which is an appropriate description for a company that had went from an idea to a tiny startup to a massive global company with over 20,000 of employees in less than a decade.

The worldwide operations of the company was complex and intense, and much of the command and control radiated from the center of the of Uber’s San Francisco headquarters. In the middle of the main floor, built from gleaming surfaces of glass and metal, stood the War Room.

To many, it was a big mystery – the War Room didn’t share the normal naming convention of city names where Uber operated. It couldn’t be booked for meetings as the others could, and was sometimes attended to by security guards. That’s because it wasn’t a normal meeting room. Many companies (inside and outside of tech) have the notion of “war rooms” but they are typically conference rooms converted temporarily to dedicated use by a product team that works intensely to tackle an emergency project, and after the situation is resolved, is quickly converted back into normal use. For Uber, perhaps appropriate to its unique needs, this War Room was not temporary at all – it was built to operate 24 hours, around the clock. It was built as a huge, permanent room with dark wood walls, multiple flatscreen TVs, a large conference table that could fit a dozen people, with additional sofa seating. Red digital clocks gave the current times in Singapore, Dubai, London, New York, and San Francisco. Given the company’s global footprint, there was almost always some kind of emergency situation somewhere in the world that needed attention, and this was often the room where it was dealt with.

That December, the emergency was in San Francisco, the company’s hometown.

Scheduled to start at 7pm and run into the night, the urgent meeting was booked on everyone’s calendar as “NACS” – which stood for the North American Championship Series, an oblique reference to its agenda focusing on operations, product roadmap, and competitive strategy in the top markets in the US and Canada. This meeting was a key mechanism for the CEO of Uber, Travis Kalanick – called “TK” within the company – to review the entire business, city by city.

A small group of about a dozen executives and leaders attended the meeting, including myself and the heads of finance, product, and critically, the RGMs — short for “Regional General Managers.” The RGMs ran the largest teams at Uber, constituting the on-the-ground Operations city teams that engaged with drivers and riders. The RGMs were thought of as the CEOs of their markets, holding responsibility for revenues and losses, the efforts of thousands of Ops folks, and were always closest to the trickiest problems in the business. I was there to represent the Driver Growth Team — a critical team responsible for recruiting the scarcest asset in the entire business, Uber drivers. It was a big effort for Uber — we spent hundreds of millions just on driver referrals programs, and nearly a billion in paid marketing. Adding more drivers to the Uber network was one of the most important levers we had to grow the business.

The weekly NACS meeting opened with a familiar slide: A grid of cities and their key metrics — tracking the top two dozen markets. Each row represented a different city, with columns for revenue, total trips, and their week-over-week change. It also included operational ratios like the percentage of trips that hit “surge pricing,” where riders had to pay extra because there weren’t enough drivers. Too much surge, and riders would switch to competitors. Uber’s largest markets, New York, Los Angeles, San Francisco were always near the top as the list, representing billions of annual gross revenue each, with smaller cities like San Diego and Phoenix near the bottom.

TK sat closest to screen, dressed casually in a gray t-shirt, jeans, and red sneakers. At the sight of the numbers, he sprung up from his chair and walked up close to the screen. He squinted, staring intensely at the numbers. “Okay, okay…” he said, pausing. “So why did surge increase in San Francisco so much? And why is it up even more in LA?” He began to pace up and down the side of the War Room, the intensity of the questions increasing. “Have we seen referral signups dip in the last week? How’s the conversion rate in the funnel going? Were there a big events this week? Concerts?” Folks in the room began to chime in, answering questions and raising their own.

A network of networks
It was my first year at the company, and although many companies have weekly reviews, Uber’s were different. First, in the discussion about each city, the level of detail surprised me. For San Francisco, the group began to discuss the surge percentages in the city’s seven-by-seven versus East Bay, versus the Peninsula. This was a senior group of executives, but the granularity and level of detail was incredible. But this was a requirement to run a complex, hyperlocal network like Uber where supply and demand went down to popular neighborhoods and frequent “lanes” — like Marina and the Financial District — which tended to be poorly served by other transportation options.

In the weekly dashboard, each row represented a city — yes — but more importantly each city was an individual network in Uber’s global network of networks that needed to be nurtured, protected, and grown. It was deeply and uniquely ingrained in Uber’s DNA to talk about metrics at the hyperlocal network level. In my several years there, it was unusual to ever hear about an aggregate number — like total trips or total active riders — except as a big vanity milestone at a company all-hands. Those aggregate metrics were regarded as mostly meaningless. Instead, the discussion was always centered on the dynamics of each individual network, which could be nudged up or down independently of each other, with increased marketing budget, incentive spend for either drivers or riders, product improvements, or on-the-ground operational efforts.

The NACS meetings were used to evaluate the health of each of the networks and the global network as a whole — a central means of accounting for the 20 or so cities that represented the majority of revenue to the company. Furthermore, it was important to go even further in granularity and break the network into the two sides, both the rider side (demand) or the driver side (supply), to make sure each side was healthy but also that they were in balance with each other. Too much surge, and riders stop taking trips. Too little surge, and drivers start to go offline and head home after a long night.

The slides continued. Several of us on the NACS team, including myself, had been working on a hypothesis over the past few days. Ops teams had reported seeing large increases in driver referrals by our primary US competitor, Lyft, over the past few weeks, which was causing drivers to switch over in droves. Driver referrals typically structured as a give/get incentive — give $250 and get $250 when your friend signs up to drive. In conjunction with a dramatic rise in demand during the holiday season, it was causing a big undersupply of drivers in the key competitive markets on the West Coast, primarily SF, LA, and San Diego. For riders, this resulted in a terrible experience — if you request a ride, it would take far longer than usual, sometimes twenty minutes, which meant more riders were canceling their requests. They might even decide to check our competitor’s pricing and service level, and book there instead. These cancels were frustrating for Uber’s drivers, who might have already driven for a few minutes. Piss them off too many times, and it might cause a chain reaction as they’d have even more incentive to stop for the night, switch off to a competitor’s network.

TK grew more intense and agitated as the hypothesis was presented. “This is not good, guys. Not good.” He exhaled deeply. What was the right solution? With the years of experience from operating these networks, it was likely that one solution would be in quickly rebalance the sides of the market. The right solution would need to start on the supply side, to grow our base of drivers quickly and lower ETAs and the cancel rate, and that meant a driver incentive. “What if… we did a $750 / $750 referral bonus here in SF, LA, and San Diego?”

This would be a big move, a far bigger number than had ever been thrown out. But SF, LA, and San Diego needed the help. These were some of the most competitive markets that would need to be quickly rebalanced with more supply. TK looked around the room, pausing, and then answered his own question. “Yeah. That would get their attention. That’ll wake them up!” he said, smiling and nodding.

Others were not so quick to jump to incentives as the solution. The past year had been good for Uber in the US, turning it into a cashflow positive area as the competition in the new China business simultaneously generated both incredible trips growth as well as severe losses. Uber was in a vicious fight with Didi — its Chinese rideshare competitor — burning on the order of a billion dollars a year primarily because of incentive spend. We started to bat around other ideas, from improving how to display ETA estimates as well as ways to discourage riders from canceling. There were other ways to rebalance the various networks without using incentives, which is a powerful tool but not the only one. The conversation went in circles, and TK grew visibly frustrated.

TK paced around the room again. “No, no! Look, guys. Our network is collapsing. We need to stop the bleeding… now!“ He chopped his hand into the other. “Let’s do the other stuff and get it on the roadmap, but let’s get this email out over the weekend. Who can help me put it together?” This decisiveness was informed by years of fierce in-the-trenches competition — companies like Flywheel, Sidecar, Hailo, and many others that were vanquished — driven by lightning fast responses in situations like this. The Uber team monitored and responded to the health of their local city networks with speed and precision. And with that, the next step was clear.

The RGMs agreed to own it, and I would work with my team — which was accountable the product/engineering side of driver referrals — to make changes to the structure and amounts. We committed to ship the changes before Monday. We took note of a number of other follow-ups due from the meeting, and we all decided to reconvene the group again next week. It was Friday and almost 10pm, and many of us had been working since early morning to prep for this meeting. I walked home, just a few short blocks away in the Hayes Valley neighborhood of San Francisco, and started my “Netflix and email” routine to close out the day.

This was my first experience with the North American Championship Series, and it turned into a weekly briefing, usually Friday mid-morning. But sometimes it got scheduled at Tuesdays at 9pm, or Sundays at 2pm, when that was the only way to get everyone together. Although NACS was just one part of my role at Uber, it quickly became one of the most educational in how to think about starting and scaling network effects. For a multiyear span, I was lucky to embedded in this critical team that operated Uber’s biggest markets. Each week was different. At the NACS meetings, we shifted our attention nimbly each session from network rebalancing on the West Coast, to prioritizing product features to increase revenues, to launching new regions, and everything in between.

Uber was already hitting its stride when I joined but I had a front row seat to the team that took grew the business to 100 million active riders in 800+ markets worldwide, and $50B in gross revenue. It was an incredible experience, and am proud of the work that we did there. It didn’t happen automatically — there were tens of thousands of people working hard to deal with network dynamics in hundreds of markets around the world, and we learned all the hard lessons from competing with fearsome local competitors who have their own strong network effects too. I’m lucky to have been at Uber during a hypergrowth period, where I joined just at the base of the hockey stick curve when it was well under a billion trips, but saw it 10x over the next few years:

My time at Uber was an unforgettable experience. I got to see a startup scale to tens of thousands of employees, millions of customers, and billions in revenue. I saw new products start at zero and then rapidly scale up to dominate the market. It was a deeply educational journey, one that created many lifelong friendships — including people I still talk to every week. But by 2018, it was time for me to move on. The company had a tumultuous few years, a complete changing of the guard, and a new set of priorities that were less entrepreneurial than in the past. I wanted the opposite of that, and for my next chapter, I decided to go back to my roots: Working with entrepreneurs to build the next new thing, but this time, as a venture capitalist.

Foundational questions
In 2018, I began a new career after Uber, as a startup investor at Andreessen Horowitz. Started a decade earlier by entrepreneurs Ben Horowitz and Marc Andreessen, the firm made a splash when it launched, quickly making a series of notable investments in startups including Airbnb, Coinbase, Facebook, Github, Okta, Reddit, Stripe, Pinterest, Instagram, and others. The firm was built around a philosophy of hands-on operating expertise — this fit me perfectly as I would parlay my lessons from Uber into picking and building the next great technology startups.

Rejoining the startup world, this time as an investor, let me tap into a network of relationships and knowledge built over a dozen years in the San Francisco Bay Area. Pre-dating Uber, I had been writing and publishing nearly a thousands essays on topics like user growth, metrics, viral marketing — along the way, popularizing tech industry jargon like “growth hacking” and “viral loops.” My blog was would be read by hundreds of thousands, and due to this as well as the natural serendipity of the startup ecosystem, I came to become acquainted with a broad community of entrepreneurs and builders. I would come to serve as an advisor and angel investor to dozens of startups, including Dropbox, Tinder, Front, AngelList, and many others. All of this, combined with my expertise from Uber, would be the foundation to launch my career in venture capital.

Everything was different in the new role. Rather than commuting to Uber’s offices in the chaotic center of San Francisco, instead I headed to the firm’s idyllic offices near Stanford University. The a16z offices combine culture and invention — its hallways lined with artwork from Rauschenberg, Lichtenstein, and contemporary artists, while its conference rooms are named after great inventors and entrepreneurs like Steve Jobs, Grace Hopper, Ada Lovelace, and William Hewlett. The work was very different from Uber’s day to day as well — rather than going very deep into one sector, like rideshare, instead my purview was extremely broad.

Every day I was meeting with entrepreneurs to talk about their new ideas. In a given year, the firm might see thousands of startup ideas, many of which are new kinds of social networks, collaboration tools, marketplaces, and other new products — relevant to the examples to this book. Conversations with startups begin with a “first pitch” meeting, where the entrepreneurs introduce themselves, show the product, and talk through their strategy. These are pivotal meetings, because when they go well, the startup could eventually receive an investment in the millions or even hundreds of millions of dollars. It’s high stakes.

Jargon thrives in these presentations: “Network effects.” “Flywheel.” “Viral loops.” “Economies of scale.” “Chicken and egg.” “First mover advantage.” These are some of the buzzwords and jargon that get thrown around in pitch meetings. And they are often accompanied with diagrams full of arrows and charts going up and to the right. The term “network effect” has almost become a cliché. It’s a punchline to difficult questions, like “What if your competition comes after you?” Network effects. “Why will this keep growing as quickly as it has?” Network effects. “Why fund this instead of company X?” Network effects. Every startup claims to have it, and it’s become a standard explanation for why successful companies break out.

But with all of these discussions and pitches, I realized I was getting confused, and I wasn’t the only one. While “network effects” and its related concepts were often invoked, there was no depth to the idea. No metrics that could prove if it was really happening or not.

In my work with startups, and after a decade and a half of living in the San Francisco Bay Area, I’ve heard “the network effect” used a zillion of times in conversation. Sometimes over coffee, in meetings, or in investor discussions, but the concept was always discussed at a superficial level.

So how do you hear something thousands of times and still not quite understand it?

I argue that we don’t understand network effects well because if it were a straightforward concept to understand, we would be in strong agreement on which companies have network effects, and which ones don’t. We would know what numbers to look at to validate it was really happening. And we’d have a step-by-step understanding of how to create and build up network effects. And yet we don’t. And it bothers me to a great degree, because it is has become a critical topic in today’s technology landscape. This is the journey that brought me to writing this book.

I began to research and to write THE COLD START PROBLEM because I found my own understanding of the dynamics of networks to be unforgivably shallow for something so core to the technology industry. The network effect is something I’ve seen firsthand at Uber, and yet I lack the vocabulary and the frameworks to articulate the deep nuances.

There’s a gap between the practitioners and the rest of the business world. For practitioners who work on specific networked products, the focus is on improving the mechanics within their very particular domains. Within rideshare, the discussion revolved around riders and drivers, reducing pickup times, surge pricing, and an accumulated set of specialized vocabulary and concepts that only apply to on-demand transportation. For a workplace chat tool, it’s about channels and discovery and notifications and plug-ins. They feel unrelated, even though both product categories have deep network effects and are both ways to connect people. There should be a set of universal concepts and theories to talk about network effects, regardless of their product category.

We need to be able to answer the basics:

What are network effects, really? How do they apply to your business? How do you know if your product has them — and which other products don’t? Why are they so hard to create, and how do you create them? Can you add a network to your product after the fact? How do they impact your business metrics, at the tactical level? Is Metcalfe’s Law actually right, or should you apply something else to your strategy? Will your network fail and will it succeed? Does your competitor have network effects, and if so, what is the best way to compete with them?

Startup advice says, all that matters is to build a great product — after all, that’s what Apple does. But why has it also been so critical to launch products in the right way? To get your product in the hands of influencers, or high school students, or aspirational technology companies — if B2B — if all that matters is the product? What’s the right way to launch, and what’s the sequence of ways to expand?

How do you build network effects in your product? How do you know when network effects are kicking in, and if they are strong enough to create defensibility? How do you pick the right metrics to optimize to achieve viral growth, re-engagement, defensibility, and other desired effects? What product features do you build to amplify network effects?

When fraudsters, spammers, and trolls inevitably show up, what’s the proper recourse? What have we seen other networks do in the past to combat the negative effects of a large, thriving network? And more generally, how do you keep scaling a network that’s already working, especially in the face of saturation, competition, and other negative dynamics?

What happens when two networked products compete — what makes one player win over another? Why did we see big networks often succumb to smaller ones? How do you launch new networks across new geographies and product lines, particularly in competitive markets?

These are the most fundamental questions we can ask about network effects, and when you search for the answers — whether in books or online — there are only smatterings of actionable, pragmatic insights though there was plenty of high-level strategy. The best thoughts came from operators, at startups and bigger companies, who have done been in the trenches and so that’s where I started the process of writing my book.

I began by conducting more than a hundred interviews with the founders and teams that built Dropbox, Slack, Zoom, LinkedIn, Airbnb, Tinder, Twitch, Instagram, Uber, and many others. I asked them questions to learn about the earliest days, when it was just the co-founders and a handful of other people trying to take on the world. I also researched historical examples spanning hundreds of years — going back to chain letters, credit cards, and telegraph networks, and tying their success to modern innovations in Bitcoin, livestreaming, and workplace collaboration tools. All of this exposed a rich set of qualitative and quantitative data which forms the foundation of this book.

I found that people were repeating the same ideas and concepts, and observed that they were recurring throughout multiple sectors. You could talk to someone who spent their career working on social networks, and find that they had ideas that were equally applicable to marketplaces. Similarly, my time at Uber made me understand the dynamics in a network of riders and drivers, which informs my view of products like YouTube and its two-sided network of creators and viewers. Or Zoom, with its meeting organizers and attendees. Dozens of these recurring themes echo throughout the industry, whether we’re thinking about B2B or consumer products.

The definitive guide to network effects
THE COLD START PROBLEM is the culmination of hundreds of interviews, two years of research and synthesis, and nearly two decades of experience as an investor and operator. It takes much of the knowledge and core concepts swirling inside the technology industry and frames them in the context of the beginning, middle, and end of a network’s lifecycle. This is the core framework I’ll describe via the major sections of this book, along with examples and hopefully inspiring you to take actionable can apply to your own product.

This is a critical topic. I’ve come to see network effects — how to start them, and how to scale them — as one of the key secrets of Silicon Valley. There are just a few dozen software products with a billion active users on the planet, and many of them share lineages of founders, executives, and investors who have unique expertise. This knowledge, in turn has been developed in the tech community over decades of building social networks, developer platforms, payment networks, marketplaces, workplace apps, and so on. This community of elite talent collaborates and cross-pollinates, switching from one product category into another, bringing all of this knowledge together. I have seen this first hand, and my interviews with founders and experts in writing THE COLD START PROBLEM further illustrated the interconnectedness of these concepts.

Based on the foundational theories of network effects, I’ve taken these lessons and put skin in the game, focusing my venture capital investing at a16z towards products that have networks at their core. I find myself most captivated by new startups where connecting people lay at the heart of the product, whether for communication, socializing, work, or commerce. I’m now 3 years into the industry, and have invested over $400M into over two dozen startups in marketplaces, social apps, video and audio, and more. I’ve found my learnings about network effects to apply widely across the industry — everything from Clubhouse, which seeks to build a new audio social app, or Substack, which lets writers publish and monetize premium newsletters for their readers. And even video games, food pickup, or edtech.

My goal is to write the definitive book on network effects — one that was practical enough, and specific enough, to apply to your own product. You should be able to use its core framework to figure out where your product is on the journey, and what product efforts are needed to drive it forward. I’ve tried to lay out the entire lifecycle — from the underlying mechanics of how to create network effects, how to scale them, and the best way to harness them — all from a practitioner’s point of view, diving deep far beyond the buzzwords and high-level case studies that have been written.

The first phase of the core framework, naturally, is called The Cold Start Problem, which every product faces at its inception, when there’s no users. I’m borrowing a term here for something many of us have experienced during freezing temperatures — it’s extra hard to get your car started! In the same way, there’s a Cold Start Problem when a network is first launched. If there aren’t enough users on a social network and no one to interact with, everyone will leave. If a workplace chat product doesn’t have all your colleagues on it, it won’t be adopted at the office. A marketplace without enough buyers and sellers will have products listed for months without being sold. This is the Cold Start Problem, and if it’s not overcome quickly, a new product will die.

This is all in the service of helping you, the reader, whether you are a software engineer, designer, entrepreneur, or an investor. Perhaps you partner with one of these companies I reference throughout the book, or are seeing technology reinvent your industry in the form of networks. Network effects are a powerful and critical force in the technology sector — as the entire economy is increasingly reinvented, it will become even more important to understand.

But let’s not get ahead of ourselves — first, what’s a network effect, anyway?

PS. Get new updates/analysis on tech and startups

I write a high-quality, weekly newsletter covering what’s happening in Silicon Valley, focused on startups, marketing, and mobile.

Leave this field empty if you’re human: