How To Improve Cohort Retention with David Lieb | Startup School

**David Lieb** (0:09)
Hi, everyone, I'm David Lieb. I am a group partner here at YC. YC has a famously simple motto, make something people want. I think it's the purest statement of the job of startup founders. And we talk about this a lot, but what gets talked about a lot less is how do you know if you've actually done it? Did we make something people want? Today, I wanna go deep on the single best way that I have found to answer this question and do it in a really quantitative way. And that is cohort retention. So high level, cohort retention is the idea of tracking what fraction of your new users keep using your product over time. Now, this isn't a new concept. There's actually a ton of content on the internet about cohort retention. It's even built into most analytics tools that you might be using. But I bet that you don't fully understand what it measures and how to interpret the numbers. And the reason I suspect that you don't understand it is that I didn't understand it in my startup until many, many years later, when it was nearly too late. I remember one specific moment, I was pitching a very prestigious VC firm for our Series A, and they asked me, hey, Dave, how's your cohort retention? And I gave them a very hand-wavy answer. And then after the meeting, I went on my computer and I googled cohort retention. And I realized that what I said must have made absolutely no sense. So I hope that I can save you that mistake here with this video. So before we dive in, quick background on me. I did YC back in the summer of 2009 with my startup Bump. Bump let you share contact information and photos with other people by just literally bumping your phones together. Bump was one of the first mobile apps to reach more than 100 million users, but it ultimately didn't work as a business. Thankfully, we did a few pivots into the photo sharing and photo management space, and ultimately were acquired by Google, where our last app, a photo management app, formed the basis for Google Photos, which I then worked on for nearly a decade. And Google Photos today serves well more than a billion users. So I've made something that a lot of people want, but I've also learned the very hard way how to know when you have not yet made something that people want. Okay, so the key insight, and the reason it's called cohort retention, is that we're going to track individual groups of new users, or cohorts, over time, instead of trying to look at the user base, or all of your customers kind of mixed together and make some inferences from that. And this gives you a much better sense of how individual users continue to use your product over time or don't. So to do this, we have to define three things. First, how we are going to isolate the cohorts.
Then, we are going to have to talk about what action we will use that counts a user as an active user or not. And finally, we have to pick which time period we want to measure subsequent usage in. So let's dive into each of these. So for cohorts, we have to pick a way to group new users into individual groups. The most common way you do this is simply by when they first used your product. So you might look at new users you acquire each week. So week one, then week two, or maybe by month. Here are all the new users that we got in January. That's one cohort. Here are all the new users we got in February. That's another cohort. That is typically the best place to start. But as you get a little bit deeper and more advanced in this, you can additionally slice by other dimensions, like country or region, or how you acquired that user, maybe what device they are using, or other characteristics of that customer. Next, let's talk about actions.
So the action is the way to define what qualifies a user to be counted as an active user in each subsequent time period. The simplest way to do this is just, did they open our app? Did they visit our site? But a better way would be to pick a specific feature that you want to track that is more correlated with them actually using your product. Maybe that's posting a photo if you're building a social network, or using a specific workflow in your B2B tool that you're building. So let's look at some examples of what popular companies might choose. Let's say you're Instagram. You would probably want to pick an action that relates to users actually seeing content in Instagram. So maybe I would pick viewed three or more posts on Instagram. And the reason I pick three is that sometimes people will open up Instagram, not touch the screen at all, not get any value and immediately leave. You probably want to filter those people out as active users. If you're Uber, you might pick completed a ride. Actually, like took a ride, ended up in a destination as an active user of Uber. For Google Photos, for my product, we chose whether a user actually tapped in and viewed a photo full screen on their phone as the best measure of whether they were an active user. Because we know that if you viewed a photo full screen, either yours or someone else's that was shared to you, you were actually getting value from the product. So, the best action to pick is one that is really correlated with the user getting real value from your product. And you want to try to filter away things where a user might be touching your product in some way, but not getting real value. Okay, so last, you have to also define the time period that you're going to measure this on. So, what I mean by this is the granularity of time that you're going to look at when you're counting whether a user performed the action that we just talked about. Typically, this should be a time period that is, you know, most connected to the desired usage of your product. So, let me give you some examples. If you're building a social app or an entertainment app, maybe your Instagram or TikTok or YouTube, you probably intend for a user to use your product every day. So, the time period in which we're going to measure the cohort retention should be daily. If you're building a utility product, maybe something like Google Photos or Uber, where you don't think a user would necessarily use it every single day, maybe weekly is a better time period to use. If you're a travel app, let's say you're Airbnb, people don't travel more than three or four times a year. So Airbnb probably would want to track something on a larger time granularity, larger time period, maybe quarterly or maybe even like semi-annually or annually. And the key here is you want to pick a time period that matches what you intend for your product. And we can talk about ways that you might screw that up later on. Okay, so now let's walk through an actual example and talk through how we're actually going to measure this in detail. So what you see here in front of you is what we call a triangle chart. Down the rows, you can see every month. So we're going to choose our cohorts by month. So all the new users in January, we're going to isolate. All the new users in February, we will isolate and so on. And then what we're going to do is track those users in every subsequent month that happens. So in this first row and first column, you see that we had 12 new users in the month of January. And what we'll do now is look in February, one month later, how many of those 12 users came back and performed our action in the month of February. And we see that it's six. And then we'll look in March, two months later, and see that we had four people come back in the month of March. The key here is that we're counting each person one time during that time period, regardless of whether they came back one time in the month or ten times in the month. Each user gets counted one time. And you can also see, if we go to the next month, we actually get five people to come back from that original group of 12 So the key is this number can go up or down. It can never be bigger than the initial group in the first month of the cohort, because that's the total number of people we're going to look at. And then we'll do the same thing for February. So February, we had 27 new users. And in subsequent months now, you can see how many of them came back each month. I like to think about this as a party. Say you're having a party, you've got your room.