This article will help you:
|
Amplitude's Personas chart groups your users into clusters based on the similarities of their event behavior. Users who behave the same way will be placed into the same cluster. It's similar to a behavioral cohort, except there's no explicit, pre-specified rule that defines a cluster.
The Personas chart rewards experimentation. With it, you can quickly do exploratory data mining analyses of the ways in which your user base navigates your product. It can help you surface similarities between user cohorts you may not have thought to look for. And it can guide you through the process of creating a comprehensive set of user personas for your product, which you can then use to drive engagement and retention.
Before you begin
First and foremost, events will not appear in any Amplitude charts until instrumentation is complete, so make sure you've got that done. You'll definitely want to read our article on building charts in Amplitude, as this is where you'll learn the basics of Amplitude's user interface. You should also familiarize yourself with our helpful list of Amplitude definitions.
NOTE: The Personas chart is only available to users on our Enterprise, Growth, and Scholarship plans.
Set up a Personas chart
If you're already familiar with Amplitude, the first thing you'll notice about the Personas chart is that it doesn't work the same way other Amplitude charts work: There's no Event Module and no Segmentation Module. There's also no Metrics Module, because the Personas chart doesn't rely on metrics the way other Amplitude charts do:
Instead, there is the Cluster Generation Module (upper left), the Cluster Count Module (upper right), and the Target Cohort Module (bottom).
NOTE: Be sure to check out our FAQ article on how Amplitude calculates clusters.
To build a Personas chart, follow these steps:
-
- In the Cluster Generation Module, choose the user cohort you're interested in from the Generate clusters from dropdown.
Amplitude populates this dropdown list with the user cohorts you've already created. If you have not previously created any user cohorts, you will only be able to choose Active Users or New Users.
When analyzing new users, Amplitude will only consider events fired during their first day as a new user. - You can limit your analysis to a segment of this cohort by filtering users based on user properties. To do so, click + where, choose the property you want to use as a filter, and specify the property value you’re interested in.
- You can narrow your focus even further by telling Amplitude you only want to include users who have already performed certain actions. To do so, click + perform, then choose the event you’re interested in.
- In the Cluster Count Module, choose the number of clusters you want to see from the ...into a total of dropdown.
NOTE: Each analysis is different, and there's no one-size-fits-all answer for how many clusters you should select. If you use too few clusters, you might find there are not enough differences between them to generate meaningful insights. If you select too many, you might find that Amplitude creates some number of invalid or spurious clusters, simply because it's unable to find 15 distinct user personas. In every analysis, you should try different cluster counts until you get a result that intuitively seems useful to you. - In the Target Cohort Module, choose your target cohort from the dropdown. Here too, Amplitude draws from the list of cohorts you've already created. You can also select from a handful of pre-set, out-of-the-box cohorts:
· [Amplitude] 2nd Week Retention
· [Amplitude] 3rd Week Retention
· [Amplitude] 4th Week Retention
· [Amplitude] 2nd Month Retention
The definitions of these cohorts depend on whether you're looking at new users or active users (including cohorts you've created yourself). For new users, they'll be included in these cohorts if they were new during the time frame of your analysis, and if they fired an active event in the week (or month) listed after they were new.
Active users will be included in these cohorts if they fired an active event during the time frame of the analysis, and then another one in the specified week (or month) following that initial event.
- In the Cluster Generation Module, choose the user cohort you're interested in from the Generate clusters from dropdown.
-
- Use the date picker to specify the timezone and set the timeframe for your analysis. Your analysis can span a maximum of 30 days.
Interpret your cluster cards
The first section of a Personas report is the cluster cards. Here, we've generated a Personas report with three clusters.
Each card consists of:
- A small bubble chart, where the size of a cluster's bubble represents the proportion of users in the specific cluster
- The count of users in the cluster, as well as the cluster's size relative to all other clusters (expressed as a percentile)
- The percentage of users in the cluster who are also in the target cohort
- An editable description field (we recommend describing your clusters)
- The option to export the cluster as a behavioral cohort
Identify and name your user personas
The cluster cards offer an excellent overview, but the real details are all found in the event table below them. This is where you'll be able to uncover the similar behaviors that hold each cluster together, and that will serve as the basis for their possible user personas.
The table displays a list of events, along with two metrics for each cluster:
- Average # of Events: The average number of times a specific event is triggered by users in cluster N.
- Standard Deviation (σ): The standard deviation from the mean of the event. Standard deviation numbers are rounded to the nearest decimal point (e.g. -0.01 is rounded to -0.0).
The table is actually divided into two halves. The top half contains events users in your selected cluster fired more frequently than average, while the bottom half contains events those users fired less frequently than average. You can sort these tables by any cluster, simply by clicking on the cluster you're interested in.
Notice that whenever you choose a new cluster to use for sorting the table, the lists of events changes as well. That's because each cluster should exhibit a different pattern of behavior within your product, which means they'll almost certainly be firing events at different rates from each other.
In the table shown below, we see that Cluster 3 is more likely to both watch a video and submit a comment than the average user.
The event table should help you answer the question: "Am I confident I selected enough clusters to do a good job of grouping my users into different personas?" If your answer is no, you should try different cluster quantities. If your answer is yes, then give your clusters appropriate descriptions and save your report.
Hide events
If there are events you still wish to track but do not want to visualize in your Personas reports, you can hide them in the Govern tab of your account. We cluster over the top 100 events, so if you mark one of those events as hidden, it will no longer count in the calculation.