Gustaf Alströmer - Growth for Startups

My name is Gustav. I'm gonna give a talk on growth for startups. This is gonna be for some of you guys, not super relevant right now because you might not have launched and thinking too much about growth when you're having a launch isn't that relevant. But for those of you that have launched, this is hopefully going to be a good talk.

So I'm going to cover three different things today. First, I'm going to talk about product-market fit and retention. The reason that that relates to growth so much is because working on growth before you have product-market fit and good retention is not a good idea. The second thing I'm going to talk about is growth channels and tactics. These things definitely apply after you have launched and often after you have a good product-market fit. You've found something that people really want and then you want to scale it up to the larger world. Lastly, I want to talk about how you make decisions when you have several people on your team. You want to start redoing things and you're not really sure exactly if you're making the right decisions or not. This is also something that applies when you're a little bit bigger.

So my background, I learned most of these things I'm going to talk about at Airbnb. I worked on the growth team for almost five years, from where it was two people until over 100 people on that team. This is the team back in 2015. Most of the lessons I'll talk about today are things I learned there. Most of you are going to be somewhere on this line. Most startups don't have product-market fit. Founders tell themselves that they do and they try to convince themselves that this is working. But the truth is for most companies, it's not working.

So that means you're going to be somewhere on this line. People also have this idea that if I launch my product, it will work. Somehow, it's going to work if I just tell the world that I built it. Unfortunately, that's not the case. The world is a really busy place and there aren't really lots of people waiting for you to launch your product. They're not standing there and they're not going to try it the moment you launch it. That is, unfortunately, not the truth.

For many people who have never thought of these questions before—how do I reach the world—this actually comes as a surprise. People have been used to working in big companies where this is not a problem. People are used to going to school or other areas where this is just not an issue. In this case, when you launch a startup, it's all down to you and this is going to be a problem in the very, very early days.

There's a great article that I recommend for you guys to read on this. It's called "Doing Things That Don’t Scale" by Paul Graham. He wrote that six years ago. It is about the early days of the Airbnb story. The thing that's really important about this is, as a founder, you need to keep two different skill sets in mind as your company grows. In the beginning of your company, you're going to do a lot of things that don't feel right, that don't feel natural to you because it's not the kind of thing that you learned in your previous jobs or in school.

It's just like the most kind of physical or real things that you have to do that you aren't going to be relevant later on. But later on, as your company grows bigger, you're going to be doing a lot of things that are things that relate directly to software and other things to scale your company. So these are two skill sets you have to keep in mind at the same time.

In NYC, we have this thing where we tell companies that you just launched; you’ve got to do things that don’t scale. We’ve got lots of these MBAs that went to school and said, “Well, this idea does not scale,” standing outside of this store or sitting in this elevator to sell people something. That certainly doesn’t scale; correct, that does not scale. But that is where everyone needs to start. If you went to school and you learned you should only work on things that really scale, you're gonna have to unlearn that skill because when you start your company, the most important thing is going to be to do things that don't scale.

So if you get comfortable with that idea, this is the early days of Airbnb. Sometimes in 2009, there were just a few people. The article I mentioned earlier, "Doing Things That Don’t Scale," tells the story of the first year or two of Airbnb when the founders came to YC. They had spent almost a year trying to get Airbnb off the ground, but it didn't really work.

This was the first version of the Airbnb website: airbedandbreakfast.com. In fact, the website itself didn’t really speak to what the company does. It was started as a website to offer air mattresses to people that visited design conferences. They had to navigate their way to find the place to where Airbnb is today. Whenever they joined YC, the first question they got from Paul Graham was, "Who are your users?"

At the time, the site looked something like this: you click on a listing and you had three different pieces of information. You had a photo of the host, you had one photo—in this case of the building from the outside—and then you have one map of where that place was. Now at the time, the only comparison to what a site like this would look like would be Craigslist. Craigslist wasn't a lot better than this, so at least they met that criteria. But it wasn't something that would make Airbnb take off. They didn't really have the product that would make everybody take off.

The things that were missing were: Is this a good listing? How does this listing actually look like? Can I trust the host? Lots of things that were missing in that early product. How do you learn that? The way they learned that is they went and talked to the hosts on their first week in YC. Paul Graham told the founders of Airbnb, "You guys go meet your hosts. Where are your hosts?" Most of our hosts are in New York; we don't have that many, but most of them are in New York.

So they flew to New York undercover; not on the cover! They claimed to be hired photographers for Airbnb. So when they met with all the hosts, they said, "Want to come by your home and take photos?" They didn't say that they were the founders because that made the company sound much smaller. They came and met with a host and, while one of the founders was taking the photos of the listings to make this look a lot better, the other founders sat down with the host and asked them questions about what challenges they were having with the product, like: What are the things that are not working? Can you show me how you use the product?

By doing that, they got for the first time to meet the people that were their customers, which they really hadn't done before, and they got to see how they used the products. That's doing things that don't scale, and that is nothing that scales. You can't go and fly to meet every single one of your customers, but when you start doing that, you will learn things that you can't learn sitting in front of a computer.

So they learned that this payout thing didn’t work, or there was a big UI bug on this page that didn’t work on Internet Explorer. All these things that you can't learn sitting in front of your computer. They went back to San Francisco, back to Y Combinator, and they sent an email the morning after. They said, "Here are all the photos we took of your house; they're now up on airbedandbreakfast.com. By the way, we fixed half of the bugs that you emailed us about or we fixed the bug that you told us about yesterday."

That made the hosts love them, and those hosts became the reason Airbnb eventually took off. Doing things that don't scale, fixing the product, making the product work for the early hosts became the backbone of the early days of Airbnb. So the lesson here is that founders are the ones who make startups take off.

The founders—you guys—are the ones that make the startups take off. You're going to have to do unconventional things. You're going to have to do things that don't feel right. Certainly, you're going to do things that you didn't learn in business school, and you're just going to do the things that are needed.

And this is basically what the YC batch is about. When someone joins YC, we're going to be like, "You’re going to launch because that's the most important thing you do right now." But once you've launched, it's like, "How do I get users?" You've got to figure out how to do it, and it's different for every company. For many other companies, that means sales. For other companies, it means doing things that don't scale. Typically, people start with their friends, and then friends of friends, and then hopefully you get one step further—the people that are now not your friends and friends and they're going to give you true opinion about your company.

Those are the people you're going to have to reach early on. It doesn't really start with, "I launched my website and I put up Google ads," or "I launched my website and somehow it's being discovered." That's not how companies get started; that's how they end up much later, but that's not how they get started.

There's only one way to grow when you're really small, and that is doing things that don't scale. All right, next topic, I'm going to talk about product-market fit. This is a terminology that probably most of you have heard of. This is a thing that's been hard to measure or hard for people to say, "Do I have product-market fit or not?" A lot of people like to tell themselves that they do have product-market fit.

It's this thing that we throw around as a way to say my product is great, so now I have product-market fit. I would argue that there are some ways you can measure product-market fit and there are many ways that you can't. So let's talk about the one thing that I think is the best way to do that.

I think that the best way to figure out product-market fit is to use data—unbiased data—to understand if you basically have made something people want. The two ways that I do that when I start: the first way is I try to figure out what is the metric, the data point that represents the value of your company; that's the first thing I do. The second thing I do is try to figure out how often should I really be doing that. A great example might be startup school. The metric here is like, are people showing up to the video talks at startup school? How often is that? It's every week. All right, that's pretty easy, but most companies can be defined this way.

Let's give some examples. In Airbnb, what is the metric that represents the value? Well, it's the bookings and the stays; it's not the searches. Search is not that; it's going to be the bookings and the stays. When I travel, I’ve experienced the value, so now I know what it is. About how often do people do this? Well, travel is actually mostly an annual thing; you don't really travel every month. Most people don't do that, so when we were measuring retention at Airbnb, we were looking at annual.

Let's look at Instagram. What's the expected use case of Instagram? It's basically just coming back to Instagram. Most people are not expected to post photos every day; it's just going to be coming back to Instagram and viewing photos. That's what most people do, and that's fine. That's actually what they want. They want some people to post for sometimes, but most of the time just coming back is good enough. How often? Probably every day.

Let’s think of a B2B company like Gusto. For Gusto, the most valuable thing that they do for their customers is their own payroll and they pay out money to employees—they're the employees of Gusto's customers.

So how often do you run payroll? Well, it depends—probably every bi-weekly or monthly. By measuring these two things, how many people am I running payroll for and are they continually running payroll with me, that’s probably the best way to figure out if people enjoy using Gusto, if they're going to switch to some other payroll provider.

And finally, Lyft. You might want to think it's rides here. The rides are the best metric here; it’s actually riders—like the people that are taking the rides—are the ones that matter because it's the individuals that we want to measure here and we not necessarily want to measure the action that they take. And that's probably weekly or monthly.

So now we have these two metrics; we have a bunch of examples of those companies. Let’s put them in my graph. One piece on the graph is going to be the metric and the other one is going to be the time window. So every single time window we can put some percentage of those people on the graph. So let’s give an example.

On week zero, in the case of Lyft, you had 100 drivers. So what do I mean by that? I basically mean that if I had, like let’s say 10 riders this week, the rate would lift; they would be calculated on the week zero. Now how many of their riders that I had last week are now traveling with Lyft this week? That is your week one number and the week two number and the week three number.

Now why is this important? Because we're trying to measure repeat usage. Repeat usage is the best, most unbiased way to figure out if someone is liking your product. It's more true than what they tell you. They might tell you things, but what they do is going to be the most important thing. So most companies can be defined this way, even if you have a B2B contract.

Companies that do annual contracts measuring—what do people do with my product—could be a really good way on a regular basis. So even if I pay for Gusto on an annual basis, which they don’t do, measuring the activity—are you using Gusto on a regular basis, let’s say bi-weekly and monthly—is the way to figure out if people are actually using the product.

So most of the ideas—even your B2B or consumer products—could be plotted on this line. Now why is this important? Well, if you're ever going to raise money, this is a graph that investors are going to ask for. Like, "How much retention do you have? Are people actually repetitively using your product?" Those are things they're really curious about because they know there are other metrics that you might have that don't matter.

This is a sign of a bad product. Basically, every single week after I started using this product, fewer and fewer people continued to come back and use the product. So this graph can be plotted and basically show that this wasn't a good product. This, however, is a good product. Every week, it eventually flattens out and the people that stop using the product stop using the product, and eventually here at week 8, 9, 10, we have a flat line of people that continue to use the product every single week. That means that they are retained; you have product-market fit for those users for this product.

So I'm not going to ask you these questions, but here are two examples of two companies that I would argue have product-market fit. The first one here has thirty percent after two months and twenty-one percent after twenty months. This is pretty good. You kept a fifty users 21 months later or twenty months later. DoorDash has a monthly retention of 20 two years later, a year and a half later. Here's another company, more like a B2B company—80% retention after one month and then 30% after 60 months. It’s really good. This is a really good product, very sticky; people like this product and they don't stop using it. It's GitHub.

So retention is the best way to measure product-market fit. Let's talk about specific things you might want to…that some people think are a better way. I'll argue that they're not. So here are some worse ways to measure product-market fit: Net Promoter Score. Why is it not good? Well, you can just Google the best products and best companies in the world—they all have bad Net Promoter Scores. Like the iPhone, Apple, all of them have bad Net Promoter Scores. It doesn't necessarily correlate with good products; it correlates with perceptions of companies.

Surveys. The problem with surveys is they are going to be biased. So if you ask your users, you're going to have some level of bias there. There are good ways to use surveys to improve your products, but it's not going to be the best way to figure out this metric. There is one cool question you can ask users, which is, "How would you feel if you could no longer use this product?" Sometimes this works; it can give you an idea, but I wouldn't do it instead of attention. I would always try to find a way to measure retention.

All right, so what are some bad metrics for product-market fit? These are not the kind of things you want to throw around as evidence for your product working: registered users—really bad; does not say anything about repeat usage or if they liked you or not. Visitors—also bad; does not say anything about whether your product is going to be valuable. Conversion rate; we have this conversion rate of visitors to something else. Well, that doesn’t really say much because you don’t know what people you’re converting; you don’t know who they are. So this does not say much about market fit either.

And finally, something that should be a paid product you’re giving away for free is not a good sign of product-market fit. You want to figure out if people are willing to pay for it because if someone says, "I love this if it's free, but if it costs money, I’m not going to use it," that's pretty bad. Then they’re not going to work out for you. So you want to make sure that the people that are doing something like this on this graph—if it's expected that you pay for a product, they should be paying for it.

All right, next section! Let’s talk about growth channels and tactics. This section really applies if you have product-market fit. If most of the people that come to your product go down the drain right away and they never come back, this section doesn’t matter. Like, why would you work on trying to get more people to a product that if no one is using your product anyway? If most people are just churning and they try it once and then they're going to come back—don't work out this stuff; wait with this stuff until you have some people that care about your product.

You can try to use some of these channels to reach those people specifically. There are really two ways that you can grow at scale. When I looked at that team, they saw the photo of the Airbnb team. They worked on two things: they either worked on what I call product growth or conversion rate optimization. What this means is you typically have engineers, designers, data scientists, and product managers working on improving specific parts of your product to get more people through that funnel.

It's a good example—I'm going to give you some examples in a second—but that's basically what I defined as the first section. Most of those people in that photo were in this category; they were engineers, designers, product managers, and data scientists. The second group is what I call growth channels. Growth channels are basically platforms in the world that people tend to discover products on.

Let me give you some specific examples: Google—huge platform for new products to be discovered. Anything that you want to use that is a rare behavior in your life, Google—that's what you do. Insurance, so I forgot insurance—Google. When I find a doctor, Google. Everything you do rarely is going to be on Google, which means lots of products are being discovered on Google, and growth channels like Google are an extremely important one for many companies.

Another one might be Facebook and Instagram. Advertising on Facebook and Instagram is critical to companies’ growth these days. What I mean by growth channels means basically other platforms about your website or your app.

So let’s talk about conversion rate optimization. What does it mean? Every single step of your product experience is a funnel that, like the retention curve, can be measured. You can have a metric, and there was a talk about this earlier, when you build funnels. If you put a metric on every single page in your product, you will know what percent of people that make it from the first page—let's say the homepage—to the booking page.

In the case of Airbnb, we called the homepage P1, the search results page P2, and then the booking page was P3, four pages; that was the entire website. Now what's the funnel? What percent of people make it from P1 to P4? What percent? Not that many—one percent, two percent. Most people don't make it that far. Your job is to figure out how many people make it that far. Why are they dropping off? What can I do to increase that number? That’s basically what multiple teams or multiple people at startups work on.

Every single step in that funnel is going to have some kind of drop-off for some reason. They might be that the content on the page is not suited for them. Like at Airbnb, all the content speaks to millennials; I have a family—it’s not good content. I land on some other website and the content doesn’t speak to me because I’m not the right customer. That’s one example of a drop-off you can fix with content—changing the content. Another one might be I’ll land on the website; it doesn’t work because Internet Explorer is not optimized for that. It’s not optimized for that; you’re going to drop off, so you’ve got to fix that too.

There are lots of different reasons why people want to drop off. Here are some specific things that people tend to work on when they work on conversion rate optimization: internationalization. If your website or your product is international, translating the product is really a good idea. We saw that at Airbnb; I’ve seen that at Facebook; I’ve seen that at many other companies where translation is really, really important.

Authentication—most products have some flow where you're signing up. Now that flow is probably your products too; it has some kind of authentication flow. That flow is very critical, and the users are kind of vulnerable in that case because they don’t really have time for too much friction. So if it’s not working perfectly, they might just go to the next website. So make sure the authentication flow works really well. Look at the best websites in the world: look at Pinterest, look at Airbnb, look at some of those sites; they have teams optimizing these flows.

The authentication flow—copy what they do; they probably figured it out; they spent a lot of time optimizing onboarding. This is a huge effort, specifically for products that need a lot of involvement from the users to become active users. A lot of questions you might want to ask early on with a new product—the more you can onboard users by asking them questions that make the experience better, the more active and the more retained they will be.

So onboarding—lots of things you can do. And finally, purchase conversion. When you're about to purchase, a lot of things around urgency and scarcity and just user flow and UI—all of these things matter, and that’s another great example of conversion rate optimization.

So let’s talk about growth channels. So again, don’t get here until you have some good sense of this. This is something people want. The first one, like I said earlier, if this is a rare behavior—most new ideas are rare behaviors, either because they don’t exist yet or because they’re not something you do every day. We tend to go to Google to learn about rare things that we don’t do very often. So that’s why, if that is the kind of product that you have, being on Google is going to be really important.

It can be either on Google through paid marketing, through SEM, or through SEO. I’ll talk about in a second. Second: does your product already share your product through word of mouth? So some products are viral in this nature because they sound really exciting to talk about. Lyft, Uber, and Airbnb are examples of those. If that's the case, you want to make sure you focus on virality and referrals. What does that mean? You're building into your product a flow that friends can tell other friends about the product.

Referrals are a way that you can do that by giving some kind of financial incentive. Does the product get better if you have more users? Well, this is true for marketplaces but it’s specifically true for anything that’s social. So if you think of LinkedIn or Airbnb, then having more people on the product is going to make it better. So it’s going to be really important for you to get more people, and those people on your site are going to be the ones doing it. So you want to figure out a good viral loop.

So when you sign up for LinkedIn, the first thing they ask you is to invite more people. That’s because your experience gets better when there are more people on LinkedIn. Now many products do work this way, and this can be perfected, and the ones who really succeed in the world of social products are the ones that really nail this down to figure out how it is really well.

The many people that make social apps underestimate how important it’s going to be to get your friends on that product. If you can make a list of all your customers—even if that list is 100,000 or 500,000, as long as it’s not mainstream enough, they’ll be in the tens of millions—you’re probably gonna do sales. You’ll make that list, and you start counting on those people. Why make it any more complicated? Why go out and reach the world for people if there are only a few people that you really want to reach?

So most companies in YC these days, I ask them this question: "Can you make a list of your customers?" Yeah, right? Make that list! Start listing them out. Who are the people decision-makers in those companies you’re trying to sell to? These people—make the list, email addresses, phone numbers—try to figure out how to reach them, but start by making the list. Don’t make it complicated by going out in a world where most people aren’t going to be relevant for your product.

And finally, this is a channel that nowadays is bigger than it ever has been and more important than it ever has been, which is: if you look at how the entire world of startups has changed in the last 10 years, more and more of them are turning more money and therefore getting what’s called a higher LTV—high lifetime value. By getting a higher lifetime value, you're enabling the ability to buy paid advertising. If you don’t have people paying for your product, where you’re making money from your product, you should not be spending time on online marketing.

Now the truth is that most companies these days are charting for the products they are making money from, and therefore they spend money on online marketing. If that’s true for you, this can be an extremely powerful channel. The biggest mistakes founders make is to start working on online marketing when they don’t have people paying for the product. Here’s an insight you probably didn’t think of: most really big companies didn’t use all of those channels; they used one or two channels.

Think of TripAdvisor. How does TripAdvisor get big? SEO. You guys type in something on Google, you land on TripAdvisor, and that's how you found this website. Most companies have a setup where there’s going to be one or two channels that really matter. If you think of Pinterest, SEO is the real way how Pinterest is going. You type something on Google, they’re already accessing the Pinterest board for that—you’ll land on that Pinterest board; that's how they acquire new users.

All right, I'm going to give some specific tactical advice on some of these channels. The first one I talk about is referrals and virality. So referrals are word of mouth. If word of mouth is a strong driver of a product, then referrals are going to be one way that you can amplify that word of mouth. How do I define referrals? A financial incentive to tell your friends about the product—this is my definition.

This is the Airbnb referral product: you give someone forty dollars to sign up for Airbnb, and when they do, I get twenty dollars. Pretty simple concept. We have that on the website and on the mobile app. Now that's actually more complicated than you might think. This entire product funnel has multiple steps in that funnel. I'm not going to go into detail here, but if you think of a referral product, it’s not just as simple as throwing that offer out.

That’s probably what you want in the very beginning, but once you have a referral part, you want to start measuring each of these steps—like what is that referral offer? And people go to that page: how many people send invites? How many of those invites are being clicked on? How many of those people sign up from those invites? How many of those people that sign up end up booking? Each one of those steps is a step in this funnel.

Let’s talk about one specific step: the referrals email invite. We would spend a lot of time optimizing this step because there were lots of people getting the referrals email invited at Airbnb. So what are the things you can optimize on the referral invite email? First, who’s the sender of the email? If it’s just Airbnb, I probably never heard of Airbnb with the first time I get this email. But if it’s Gustav that sends the email and I send it to my friends—they have heard of me; that’s the reason to open the email.

So people open the email—clear value. What’s this email about? Many emails you start with text. Don’t start with text; just have the clear value prop at the top. Why should I care about this email? In this case, it’s extremely simple: “Gustav sent you 40 for your first trip.” That sounds good! What is that about? I’m going to read about it.

When do I have to care about this? By this date in the next month, so I can’t just leave this email and never open it again. I have to do it right now. What do I have to do here? Well, I could sign up, which is an undefined thing that I can do sometime in the future. Or I can do what we did here: accept my invitation. This sounds more exclusive; it sounds like something that is just for me. It doesn’t sound like something I can do anytime in the future. And finally, here’s some social proof from this email: this is me! I live in San Francisco! We can reveal that I actually been a member of Airbnb since 2009, and we can reveal that as well.

Let’s talk about paid growth. Each of these sections—referrals, paid growth, SEO—could be a presentation on its own, so it’s impossible for me to go into deep details on this. But if you're determined that you have product-market fit, you want to grow one of these channels. And this is the channel you want to go deep on. You’re going to have to go really deep on it because being really good at one of these channels requires a lot of work.

So there’s lots and lots of stuff online about how to get really good at one of these channels. It doesn’t really make sense to get good at all of them because most of you won’t really need all of them. The number one lesson in paid growth, i.e., online marketing, is to not do it unless you have revenue. This is the most common mistake that founders make—is that someone starts buying ads for products, and they will never be able to pay them back. Don’t do that.

The next thing you want to figure out is what’s called CAC—customer acquisition cost. How much does it cost to acquire a new paying or a new valuable customer? Someone’s giving you value back. Many of the advertising tools like Google and Facebook have a very clear system for how they calculate this, and once you start running ads, they’ll start telling you what the cost is going to be.

Next is going to be that your revenue or projected revenue from this user is going to have to be higher than the CAC—higher than the cost. Very simple; otherwise, you can’t do this. So how do you know? This is the common question you get early on in paid marketing. Well, it seems like in eight months it will be higher, but not in the first month.

Well, you can’t take all your money and spend it on something that you have no clear certainty of is going to happen in the future, so you're going to have to either wait eight months or look for early indicators that your hypothesis about the value is going to be stronger. The best thing a startup can do is don’t wait eight months; just have a much lower target on what your CAC is going to be—maybe one month, two months, three months, first transaction, something like that. That’s a much better way to do it.

The main channels for online marketing these days are going to be Google, Facebook, and Instagram—that’s pretty much it. Let’s talk about search engine optimization. This has changed a lot in the last couple of years; it’s very competitive. And what changed is there used to be millions of websites; each would rank for tens of millions of keywords.

Now what’s changed is that the really big companies are starting to get really good at ranking for all those keywords. So a Pinterest or a TripAdvisor might rank for every single travel keyword you can imagine. That’s hard for small companies. What that means is if you’re going to rely on search engine optimization to grow, you’re going to have to be as good as a Pinterest or a TripAdvisor eventually—not right away, but eventually. It’s so competitive to win in this grand, large world of SEO.

When you get started, you can think of it this way: SEO is basically a zero-sum game. You’re competing against others, so what you do in SEO is going to matter and what you compare to others. The second thing is that the keywords that people search for are changing constantly. So if you’re building something new, let’s say ASMR, I think was a thing that came up recently—lots of companies able to rank for that because it’s a new keyword. There weren’t websites built 10 years ago that rank for that because the thing didn’t really exist.

All right, let’s talk about SEO and how it works on the technology side. This is their search results page; this is what you and I see when we go to Airbnb. This is what google sees; google just sees text. So to be good at SEO, you need to understand what text am I showing to Google so Google can understand what the website is about. If people can’t understand what your site is about, it’s not going to rank.

What are the two main levers for SEO? The first one is going to be things I do on my page. So for example, what’s the title of the page? Can Google read the page? Does the page throw errors? What specific page in my keyword am I trying to rank my page for? Well, start with the keywords—do some research and see what people are searching for.

How many people are searching for ASMR in the United States per month? Maybe I want to try and rank for that keyword. Well, I’ll build a website just trying to rank for that keyword; start with Google. Don’t start with your own content. If you don’t know exactly what people are searching for, you’re going to start doing some research.

The second thing is the thing you can't do that much about, which is called off-page optimization or domain authority, which basically means how valuable does Google perceive your website to be in the grand scheme of all websites? The more inbound links you get from press, the more links you get from all kinds of people that are also authorized—like have high authority—the more valuable your website will be in the eyes of Google, which means it will rank you higher on some of the keywords you’re trying to rank for because it will compare you to other websites and see if they seem more or less authoritative.

I’m not going to details here, but that’s basically how Google works. If you're curious about this, you can Google Page Rank and go to the Wikipedia article on Page Rank—it basically will explain sort of like high-level how Google works.

Final section, I'm going to go through this one a little faster. Most of you guys don’t have to focus on A/B testing at all; it won't matter for a long time. It is a great decision-making tool later on. Here’s the situation the startups tend to get into: I want to launch a new homepage; I want to launch a new design. I did, and the numbers went down. What happened?

It’s a really hard problem to launch something new and sort of like just look at the metric over time. Don’t do that; there’s a better way. First, before you get into that stage, you want to figure out: is A/B testing something I want to do? The best way to do that is to go to Google and type in “A/B testing calculator.” Think of the metrics that you’re trying to change here, so like visitors to some conversion metric—put them into that first link you see on Google and that’ll tell you whether it’s going to be worth doing. Most of you, it won’t be worth doing for quite a while.

So here’s the example I’m trying to give you on the website: so I want to ship a new experiment—our new design on the homepage. So let's ship it! The metric went up or the metric went down; either way, I don’t know if the website actually caused it or not. The only way for me to know if this new design actually changed the metric is if I had an alternative side of history—like two sides, two different parallel universes at the same time: one with the new design and one with the older design.

If I had that, I can tell exactly what happened. That’s the definition of A/B testing: you basically have two different parallel universes of the thing you shipped at the same time you measured the metrics that matter to you. The reason this is so powerful is it helps you make decisions at scale. What ends up happening is founders, when they get five or ten people in the company, and they launch a new design, they're arguing about what caused the thing to go up or go down.

The only way to really know is to run an A/B test to figure out what does the metric say about what it went up or down. This is hard to internalize because most people think of themselves as good product thinkers. We're talking about one thing called the "experiment review.”

This is Airbnb. How many here think that you guys have good product instincts? They usually hand—now you guys are founders; you should have good product instincts. All right, let’s see. Let us say go there. All right, so I'm gonna give you two examples. So at Airbnb, we launched a new sharing sheet for the mobile app. This was the old version, which was the native share sheet that you've seen on iOS.

You click on share and then you see a bunch of sharing options, and then we just tried this new sheet that showed more options. We call this the experiment, but it didn’t really look native. So the question was, which one was better? Well, we didn’t really know. So we launched an A/B test. We launched both of them at the same time for different users. The goal here was to measure the number of shares.

So how many here think that the control was better? How many people think that this experiment was better? And people thought there was no difference. This is quite common; this was about forty percent better for us. So thank god we did an experiment because if this was the decision-making group, then we wouldn't have made the right decision.

Next one: should we have a signup wall or not in the app? Now these are not necessarily learnings you can apply to your companies right away, but it was an important decision for us to determine. So should we have people just open the Airbnb app and go straight into the app, or should we have an experiment where you can click out and exit out the signup wall, or should we have a signup wall that is a wall that you can’t climb over? You have to sign up; otherwise, you can’t use the app.

Which one is better? How many here think that the control—not having a signup wall—is better? Raise your hand. How many people thought that the experiment where you can exit out and then sign up was better? And people thought that just like this wall you can't climb over it was better— that's a few amount of people. All right, so this is a lot better. We got 2.6 more bookings from iOS by making people sign up through a sign-up wall in the app.

Why is that? Well, we knew something about them so we can send—we can show them more personalized stuff, and when they were about to book, we already had them signed up. So they don’t have to at the time of booking go through the motions of signing up.

You want to learn why? Why does this happen? The whole point is basically when you get big enough, when you're starting to grow and you have these decisions about, "Should I launch this thing or not?" This is a really good way to do it. Practicing decisions are really hard. So using data to make them is a good way. Most of you won't have to worry about this for a while, so don’t worry about it.

Here’s the summary of my talk today; most of you need to do things that don’t scale. You are not at the place where you can think about real growth—things that growth teams do. So you have to unlearn the things you've learned at your big companies or in MBA programs and just do things that don't scale.

Secondly, you want to measure your attention to understand if you have product-market fit. There are other ways too, but that's the best way in my opinion. And third, you want to build a culture of experimentation. You want to use data, not have the loudest voice in your room decide what the best decision is, but you want to use data and experimentation to decide what is the best decision.

Probably doesn’t matter right now, but it will matter at some point. Thank you!