Create your lifecycle cohorts, learn the three ways to measure user retention, and turn all this work into clear results.
Now that you’ve done some investigation into the state of your analytics, your current metrics, and your product’s usage, it’s time to dive into the meat of this playbook. In this chapter, we’ll introduce the Retention Lifecycle Framework, our in-depth framework for improving retention based on how users interact with your product.
In the first half of this chapter, we will discuss three ways you can analyze your retention as well as how to do these analyses in the Amplitude platform. The second half of this chapter is about how and why you should look at your users in 3 separate stages—new users, current users, and resurrected users; the flow of users between these three stages is what makes up the Retention Lifecycle Framework.
As we mentioned in Chapter 1, retention is the proportion of users who remain active in your product over time.
Since there are a few different ways to think about retention, it’s important to make sure you know how your analytics platform defines it and whether this is suitable for your product. (If it’s not clear how your analytics platform defines retention, get clarity on this before going any further!)
At Amplitude, we let you choose between three different types of retention: N-Day retention, unbounded retention, and bracket retention.
Figuring out which type of retention calculation makes the most sense for you depends on your business goals and how your users naturally use your product—your critical event and usage interval (Chapter 2) should have given you a good understanding of this.
Now, let’s look at each type of retention so you can determine which is best for your product.
As we mentioned in Chapter 1, when most people discuss retention metrics, they’re talking about N-Day retention. N-Day retention measures the proportion of users who are active in your product on a specific day N after first use.
N-Day retention is well-suited for gaming or social apps, or any other type of product where you’re trying to get users to exhibit regular, repeat behavior.
The day a user is first active in your product is designated “Day 0.” This could be the day that a user first downloaded your app, the day they registered, or the day they performed any kind of action within your app—played their first song, added their first friend, etc.
For all the new users who first became active on a certain Day 0, you can calculate their Day N retention over the days that follow, looking at what proportion of users were active on Day 1, Day 2, Day 7, and so on.
For example, Day 1 retention looks at how many users returned specifically one day after they were first active; Day 3 retention looks at how many users returned on exactly the third day after they were first active; Day 7 retention looks at users who returned exactly on the seventh day after they were first active, and so on.
This means, if you were looking at Day 7 retention, a user must be active on Day 7 to be counted; it doesn’t matter how many times users came back between Day 1 and Day 6, or if they returned on Day 8. If they aren’t active on Day 7, they are not considered Day 7 retained.
The N-Day retention curve below shows the weighted averages of all of the N-Day retention numbers in a Day 0 to Day 30 timeframe.
What if you expect users to use your product regularly every week or every month instead of every day? In this case, ‘N-Week retention’ or ‘N-Month retention’ would be more appropriate. Conceptually these are the same as N Day retention.
Third week retention, for example, reflects the proportion of users who are active any time during the third week after they were first active (Week 0). Similarly, third month retention reflects the proportion of users who are active any time during the third month after they were first active (Month 0).
Unbounded retention shows you the proportion of users who came back on a specific day or anytime after that day. This type of analysis can be a better fit than N-Day retention for businesses that don’t expect users to come back and engage with their product or service on any regular cadence.
Note: Unbounded retention is sometimes called “rolling retention.” We have chosen not to use this term because you are not calculating a true “rolling” or moving average using this method.
The curve in the image depicts the same set of data-points that we looked at in the N-Day retention graph above, but this time shows unbounded retention numbers.
Notice our unbounded Day 1 retention rate is 50%, compared to only 21% for N-Day—this means that 50% of users who were new on Day 0 were active at any time on or after Day 1, whereas the N-Day retention rate only counts users who showed up on Day 1 exactly. Unbounded Day 2 retention is around 38%, which means 38% of users who signed up on Day 0 were active at any time on or after Day 2.
Put another way, unbounded retention is actually the inverse of your churn rate. By measuring the inverse of your unbounded retention, you can see precisely how many users used your product on Day 0 and never returned again.
A grocery delivery service does not expect people to use their product on a daily basis; they might not even use their service with a predictable cadence. Instead of looking at whether someone comes back exactly on Day 7 or Day 30, which is what N-Day retention would indicate, this company would get more value out of looking at their unbounded Day 7 retention—that is, how many new users return to buy groceries after their first week.
Amplitude also allows you to calculate bracket retention. Bracket retention is a more nuanced version of N-Day retention; it lets you split up your retention analysis into custom retention periods, instead of limiting yourself to a daily, weekly, or monthly timeframe.
Once you understand your users and their expected usage patterns, you can begin to use bracket retention and define your own custom intervals of multiple days, weeks, or months.
Here, we’ve set up four brackets in Amplitude:
Let’s say for this example that an active user is someone who comes back and performs any kind of activity. Then, someone would be counted as fully retained by:
In the image below, you can see what a retention curve might look like with these brackets
Each multiple-day bracket is like a bucket, and if a user is active anytime inside that bucket, they’re counted as retained. The spacing of each bucket in time is based on the pattern that you want to see your users exhibiting.
For a product that people use every three weeks to stock up on home goods, we might create buckets that span about three weeks, plus or minus a week. In that case, we don’t care whether people come back exactly on Day 1 or Day 3, or whether their second order is after two weeks or a month—we just want to see a rhythm that proves they’re getting value.
Pinterest uses one type of retention metric, which they call “1d7,” to measure how many of their new users come back and discover value in their product (for more on value discovery, see Chapter 6). This metric looks at the proportion of new users who who visited Pinterest any time between Day 1 and Day 7. The growth team also looks at retention of these users in the Day 28-Day 35 bracket, in order to know what percentage of new users are still active one month after signup.
To summarize the key differences between N-Day, unbounded, and bracket retention:
How do you figure out which type of retention to use? There’s no quick and easy answer, but it depends on a combination of your product’s usage patterns and your business goals.
Figuring out which type of retention to use is heavily dependent on how frequently you expect people to use your product. If you expect people to come back on a regular basis, like daily for a mobile game, or weekly for an exercise app, then N-Day retention is probably a better fit. If you notice that many of your users don’t have a steady usage pattern—for example, a food delivery app where people place orders sporadically, then unbounded retention may provide a more accurate measure of how your business is doing.
To start out, you can try measuring your retention via a few different methods to see which one gives you the most meaningful information.
The right retention metric should show you where you can improve and give you an accurate view of the health of your business.
By now, you’re hopefully used to seeing a typical retention curve like the one above.
The fundamental problem with a retention curve like this is that it lumps together a lot of different types of active users in one single curve.
In reality, not all active users are created equal. In order to make meaningful, long-term improvements to your retention, you need to understand your active users as they flow through different stages of being retained.
The Retention Lifecycle Framework can help you accomplish this goal.
What do we mean by a user that is “active”? In Chapter 2, we introduced the concept of a critical event—an action that you want a user to perform in order to be counted as truly active or retained.
We recommend measuring retention based on whether users come back to your app and perform your critical event, not just whether they open up the app. This will give you a more accurate view of how many users are truly getting value out of your product.
What do we mean when we talk about the “retention lifecycle?” The way we think about analyzing retention and putting strategies in place to improve it should change depending on what stage a user is at in their product journey. Active users go through three different stages of retention: new user retention, current user retention, and resurrected user retention.
These three groups make up your total active users at any given time. If you’re a daily usage product, this means that on any given day, an active user of your product will be at the new, current, or resurrected user stage.
This image maps how we think about the flow of users between these different stages of retention:
The main objective of the Retention Lifecycle Framework, and the Retention Playbook as a whole, is to get your existing new users, current users, and resurrected users to become more engaged current users.
To achieve retention that rivals the likes of Snapchat and Instagram, you have to engage differently with new users and current users, put strategies in place to resurrect inactive users, and move all of your users toward being more engaged overall.
TERMS TO KNOW:
New Users are users who are using your product for the first time.
Current Users are users who have been using your product consistently for some period of time.
Dormant Users are users who were once actively using your product and then became inactive.
Resurrected Users are users who were once actively using your product, who then became inactive for a period of time, and then became active again.
Chapters 5, 6, and 7 will cover each of these user stages in more detail, but here’s a quick high-level overview.
A lot of the existing content about improving user retention focuses on how to retain new users—things like revamping your onboarding flow or sending new user drip campaigns, for example.
This makes a lot of sense, since so many users churn within the first 7 days. But not focusing on engaging your current users or finding ways to resurrect inactive users would be a huge wasted opportunity.
Don’t take your current users for granted. Every current user has the opportunity to turn into a highly-engaged power user. Your goal for current users is to continue providing them with value and keep them coming back.
In the next chapter we’ll talk about how to cluster your users into different behavioral personas, which can help you further understand and capitalize on the value (or values) that current users derive from your product.
Dormant users are, in fact, the largest percentage of most products’ potential user pool. Many of these users are probably using a competitor’s product, so they’re high value as well. There are also numerous studies that show that it’s cheaper to resurrect a dormant user than it is to acquire one.
When you find that more dormant users are coming back to your product, it’s important to invest time into figuring out why. Did they respond to a particular winback campaign? Push notification? Did they become current users or did they drop off again?
Why does the retention lifecycle matter? Because too many products try to artificially increase their active user counts through simply acquiring new users. Of course, top of the funnel is important (if you can’t attract new users you have no one to retain) but the growth of your current and resurrected user base is what really matters for true growth.
Breaking down your active user base into new, current, and resurrected users shows you whether your product is experiencing true growth.
Let’s take a look at the example below:
Here we have a bar graph showing the total number of active users for a certain product over the course of 12 weeks. The product grew from having 6 million active users to just over 8 million active users by week 12—things seem to be going great, right?
Break down these users into new, current, and resurrected users (graph on the right), however, and it’s a different story.
New users are definitely increasing over time, but your number of current users is actually decreasing. If you take this one step further and graph the number of users who become dormant every week, you’ll see that this population is getting larger and larger over time.
Bottom line: Although you may be gaining more new users, you’re not experiencing true growth if a large proportion of users end up churning and you don’t have a sustainable, growing base of current users.
In product analytics, the broadest definition of a cohort is a group of users who share some common characteristic. To analyze your new, current, and resurrected users in the next few chapters, the first step is to create these cohorts.
Depending on what analytics software you’re using, you can either define these cohorts within that software or in your raw data.
As we alluded to in Chapter 2, your product’s usage interval is an important part of measuring your retention accurately, across all stages of the Retention Lifecycle Framework. This interval determines how you define your timeframes for new, current, resurrected users, as well as your dormant users.
TERMS TO KNOW:
A new user is a user who is in their first interval of using the product.
A current user is a user who used the product in the previous interval and the current interval.
A resurrected user is a user who is active in the current interval, but was not active in the previous interval. Also, this user was active at some time prior to the previous interval (i.e. they are not new). Note that a user can only be “resurrected” once they’ve become dormant, which is why this definition requires one interval of inactivity.
A dormant user is a user who did not use the product in the current interval but was active in the previous interval.
In Amplitude, you can use the Lifecycle feature to see your retention lifecycle breakdown and create your cohorts in just a few clicks. Lifecycle breaks out your active users into new, resurrected, and current buckets, and also shows you your churn in each time period. This example shows a product with a weekly usage interval.
When you hover over one of the bars, you’ll see a tooltip. This is Amplitude’s Microscope feature, which allows you to dig deeper into any data point and see the users and behaviors behind it. To create your cohort, just click ‘Create Cohort’ and give it a name. Repeat this for new and resurrected users.
Once you’ve created your cohorts, you can see what percentage of your active users fall in each stage of the lifecycle. Here’s an example from a product’s split in a given week:
Creating a table like this will give you an idea of where your strengths and weaknesses are. In the example above, over 70% of the active users in the time period are new users. For the product to grow sustainably, this company needs to make sure they are retaining new users well to grow their current user base each period, and not only focusing on acquiring new users.
You can also see that they have pretty high churn relative to their total active users. While they definitely want to reduce that churn, this is also a large base of users to potentially resurrect, which we’ll cover in Chapter 7.
Here’s another example:
This product has a much healthier distribution of active users. Almost 70% of their users are current users who were active in the previous period, and their churn is pretty low relative to the total number of active users. A company with a distribution like this already has pretty good retention, and can put more efforts behind top of funnel to drive more new users and accelerate growth.
To take note of your own retention lifecycle split, be sure to complete the worksheet “Your Retention Lifecycle” at the end of this chapter.
Pulse is a chart view in Lifecycle that lets you get a quick pulse on your product growth. It depicts the ratio of incoming users to outgoing users for a particular day, week, or month and allows you to see how many active users you gain for each user that churns.
The Pulse ratio is calculated as:
(# of new users + # of resurrected users) / (# of dormant users)
Note that since both newly acquired users and resurrected users increase the pool of active users, they contribute to your total user influx.
The chart above shows overall growth for this product decreasing between November 28 and December 4. At its highest point, on November 28, the pulse ratio was 2.19, suggesting that for every 2 users gained, one was lost. At its lowest point, on December 3rd, the pulse ratio was 0.36.This means, for everyone one user gained, roughly three users were lost on this day.
If you don’t use Amplitude, you can still manually calculate this ratio after splitting up your user base into the lifecycle cohorts mentioned in Section 3.3.
The purpose of this chapter was to introduce different ways to analyze retention as well as the Retention Lifecycle Framework. Whichever retention analysis you choose for your business, the framework remains the same: first get a deep understanding of how users retain at each stage of their lifecycle, and then put in place strategies to turn your active users into highly engaged current users.
In the next several chapters, we will explain the deeper nuances of new, current, and resurrected user retention. To prepare yourself for those chapters make sure you:
PRO TIP: Be careful about over-optimizing for just your power users.
Understanding power usage is important, but you can’t convert everyone into a power user overnight. As you iterate on your product, make sure to optimize for retaining the most number of users, not just the best users.
For example, say you’re an on-demand delivery app and find that purchasing at least 7 items at a time leads to significantly higher long-term retention—however, only 2% of your users actually do this action. It would be a bad idea to encourage users to buy more by increasing the minimum cart size, as this would exclude a huge proportion of your current user base. Many of them may end up churning as a result!
In this chapter, we introduced the Retention Lifecycle Framework, which breaks out your active users into three separate stages of retention: new, current, and resurrected.
Your product usage interval that you calculated in Chapter 2 determines how you define your timeframes for these stages. Remember, the definitions of each type of user are:
Measuring your current distribution of new, current, and resurrected users will help you identify your strengths and weaknesses, as well as what stage you might want to focus on improving first.
Complete the table below by recording the size of each cohort in a given time interval (equal to your product usage interval from Ch. 2). Since you’ll want to measure the long-term retention of these users, we recommend choosing a time frame that is at least 2 months before today.
For more details on how to create these cohorts, refer back to Section 3.3. If you’re using Amplitude, you can use the Lifecycle chart instead of calculating these cohorts manually.
“Pulse” is a ratio that gives you a quick check on the health of your product growth. In short, it’s a ratio of incoming to outgoing users for a time interval.
Using the numbers you recorded in Step 1 of this worksheet, calculate your Pulse ratio as:
If you’re using Amplitude, you can use the Lifecycle chart instead of calculating this manually.