Visitor Retention Analysis with Adobe Analytics

Once you’ve defined the new users(t) metric, defining the repeat users(t) is a piece of cake. What are repeat users on a given month t? Well, they essentially are all unique users that are not new users on month t. In other words:

repeat users(t) = MAU(t) – new users(t)

Or visually:

So here’s how you define repeat users(t) in Adobe Analytics:

Defining Retained Users(t) and Resurrected Users(t)

Now, notice that repeat users(t) are also the summation of retained users(t) (those that we retained from last month) and resurrected users(t) (those that we reactivated from a month prior to the last), as shown here:

repeat users(t) = retained users(t) + resurrected users(t)

Therefore, once you’ve calculated repeat users(t), you can either calculate retained users(t) first and then subtract it from repeat users(t) to obtain resurrected users(t)or, vice versa, calculate resurrected users(t) first and then subtract it from repeat users(t) to obtain retained users, as these two derived identities show:

(1) retained users(t) = repeat users(t) – resurrected users(t)

(2) resurrected users(t) = repeat users(t) – retained users(t)

I’ve tried both methods as a validation to one another and both work. In this article I will present the method suggested by identity (2) above, so I will calculate retained users(t) first and then derive resurrected users(t). Here’s how to go about it:

Step 1. Define New Monthly Segments for Retained Users

Like we did for new users, we need to define 25 new segments, one for each month during the last 24 months + current month. Here’s the example for retained users (10 months ago):

Here’s a problem though. You would expect a segment like this to give you visitors only for the full month if you go back 10 months ago from current month. However, given its very definition, Adobe Analytics will make that same set of visitors appear on both full months 11 months ago and 10 months ago. You can see this visually if you define the following metric:

If you add this metric to a free form table, here’s what you would see at the time of writing (29 October 2017):

However, we are only interested to know retained users for the exact full month that stands 10 months ago from the current month. How do we solve that?

Well, we need some sort of month selector so as to show the retained users on the retention month only (Dec 2016 in this case). That’s what we will do in step 2.

Step 2. Define Month Selectors

A month selector is nothing more than a simple metric containing 0’s for all months except for the one we want to select, which needs to contain 1 as value. We will later multiply that metric by the metric calculated earlier for retained users so we will be able to get rid of the double attribution.

Keeping on using the example of 10 months ago, what you need to do is to use the previously created date range 10 months ago in order to build a new segment that only captures the visitors that visited your website on the full month that stands 10 months ago relative to the current month, like so:

Now, apply this segment to the unique visitors metric and here’s what you get:

We are just one step away from the month selector metric we want to build. You see, the value on Dec 2016 needs to become a 1. What’s the easiest way to make that value become 1? Just divide it by itself! However, as easy as this sounds, calculated metrics only allow for operations amongst metrics or between metrics and constant values (called static numbers in AA). So, in order to divide that value by itself, you need a metric that has that exact same value for the month in question. And what is that metric? It is simply Unique Visitors. So, all that we need to do to define the 10 month ago selector is to apply this formula:

And here’s how the result looks like on a free form table:

Exactly what we wanted.

Step 3. Calculate Retained Users(t)

Putting it all together, the formula for the the retained users(t) metric is the following:

Step 4. Define Resurrected Users(t)

As explained above, you now simply calculate resurrected users(t) as:

resurrected users(t) = repeat users(t) – retained users(t)

Defining the Churned Users (t+1) Metric

Last but not least, here’s the process to define the churned users(t+1) metric.

Step 1. Define New Segments

Similar to what we’ve done for new users and retained users, we need to define 24 new segments, one for each month during the last 23 months + current month. You will immediately notice a difference though: we’ve got one month less here, and that’s due to the very nature of this metric. We can’t in fact have churned users at time t. You easily see that as you define one of the segments. Keeping on using the example I’ve used so far, say you want to build the segment of churned users (10 months ago). You would define it this way by making use of date ranges previously defined:

As you can see, while we have defined these visitors as churning 10 months ago, this segment will however contain visitors attributed to the date range 11 months ago but contains no visitors for the date range 10 months ago. Why? Because of the mere fact that the segment definition above excludes users that have appeared on the platform 11 months ago. This means that we are in fact calculating churned users at month t+1 for any given month t.

Step 2. Calculate Churned Users(t)

All new segments having been created, here’s how you then compute the churned users(t) metric:

The multiplication at the end of the formula is required to make churned users a negative number (good for visualisation). It is a bit dodgy: I wish there was a way to multiply by -1 but there isn’t.

Conclusions

While getting to this point requires a heavy lifting, it is worth the effort if you are looking to use these segments and metrics as a basis for more advanced product analytics.

Please bear in mind that the retention cycle I’ve focused on for my analysis is monthly based. Therefore, these metrics will work fine if you break them down by month. But please notice that while you can scale them up to focus on longer intervals of time, such as quarters and years, as well as scale them down to analyze shorter intervals of time, such as weeks, days, hours, etc, you won’t be able to break them by quarters or days straight away; if you want to do so, you will need to redefine them by using longer or shorter intervals of time in your data range definitions.