Historical Count unlocks a deeper level of analysis when digging into why users are retaining, converting, or engaging (or not). You may find that conversion and retention rates are quite different for a user that has made a purchase in the app for the first time vs. second time vs. third time etc. Quickly pinpoint areas of friction for first-time users in order to chip away at improving your overall North Star metric. Furthermore, you can define your best customers based on the number of times they’ve done a critical action such as completing a purchase in an e-commerce platform or playing a song in a music streaming app.
What you will learn in this article
Historical Count allow you to capture the Nth instance of a user action right out of the box. For example, since you can separate a user's first time interacting with a new feature you shipped versus the 2nd time they interacted with it, you're able to better identify points of friction in feature on-boarding. If your product was a music streaming company, Historical Count could provide the usage difference of when users performed "Play Song" event for the first time or multiple times.
This article will go over explaining this feature, as well as explaining how it is different from Behavioral cohorts.
Prerequisites
- This feature is available only to customers on the Growth and Enterprise plans.
- A completed instrumentation is necessary to have events appear in the Event Segmentation chart and all other Amplitude charts.
- [Recommendation]: If you are new to Amplitude, we recommend to start exploring the Event Segmentation chart first.
Other important things to note about this feature
- Historical Count is an [Amplitude] Event property available on Event Segmentation, Funnel Analyses, Pathfinder, and Retention Analyses Charts.
- It’s only possible to apply an Historical Count filter on the 1st event in the Pathfinder and Retention Analysis charts.
- It can be applied to multiple events in Event Segmentation and Funnel Analysis.
- There is a possibility to do 1st, 2nd, 3rd, 4th, and 5th time analysis.
- Currently, there is a possibility to only select one "Historical Count " property at a time (not possible to select 1st OR 2nd time analysis in the same event line).
- Currently, only the “=” operator is supported.
- Currently in Funnel Analysis, Historical Count works on "this order" funnels.
Historical Count definition
To proxy whether this is a user's Nth time, we have to look back at their historical data. Amplitude uses a 1 year lookback window (1 year prior to the start of the date range). So when the chart has a starting date range of March 1, 2020 - the lookback window will go back to March 1, 2019.
"First-Time" is approximated by looking to see if users in a given date range performed the event of interest in the 1 year preceding the start of the date range. If not, the user is considered to have performed the event for the first time in the date range. For example, if a user performs the "Play Song" event for the first time on February 20th 2019 and second time on March 20th 2019, and the chart is querying for data with date range of March 1, 2020 - March 31, 2020, since there is 1 year lookback the event from March 20th 2019 will be considered as the first event.
"Second-Time" is essentially the “2nd time” a user performed an event in the last year, etc..(the max Nth time analyses is 5th).
To apply a Historical Count filter, select the "Historical Count" under "[Amplitude Event Properties]" and the value that you are interested in.
In the following example, the chart is checking for Users who performed "Play Song" event with a
"[Amplitude Event Properties] Historical Count" = 1st.
How is the filter applied?
The Historical Count filter is applied “last.” For example, if the event is “Play Song” WHERE “country” = “US” AND “nth time” = “1st”, the chart is computing the user's first time performing Play Song in the US. This will not show a user’s first ever performed event that happened to be performed in the US. This means that if the user has performed their first event in Germany, but the chart is querying for events performed by users while in the USA, the first event that the user performed in US will be considered.
The "Group By" feature will apply to the first event selected in the Funnel Analysis, Pathfinder, and Retention Analysis charts. When you apply a Group By, we will show what the user's property value was at the time of performing the event for the Nth time.
Funnel Analysis
In Funnel Analysis charts, a user could enter the funnel multiple times or perform the various steps multiple times. We strictly look at the Nth time the event was performed to be consistent with how we'd regularly calculate a Funnel. For example, if we have:
Step 1 = Event A
Step 2 = Event B
If the user performs BBABAB...
We will count the second time the user performed B. We are looking at the second B and if they performed A before that second B (within the conversion window). The user did not perform A before the second B, so they wouldn't count as a conversion.
Additionally, when using Historical Count filters on the same events that happen within the same second, Users will appear as dropped off. This is because the Funnel query does not distinguish between events that happen within the same second, but the Historical Count filter does.
How is Historical Count different from Behavioral Cohorts?
Behavioral Cohorts can be used to define a group of users who did a specific action such as “completed workout” 5x’s in the last 30 days. This might be a fitness company’s definition of a recent power user: someone who completed a certain amount of activity in a specified time frame.
Conversely, Historical Count allows you to pinpoint a user’s 5th workout. So if they did a workout only 2x’s in the last 30 days, but they had also previously completed 3 workouts, the most recent workout was actually their 5th workout. This is an important distinction, because a user’s 5th workout could mark an important milestone in their overall user lifecycle: they have now transitioned into a group of users that are highly likely to retain long-term, or be a frequent purchaser. This is a different type of user than the person who completed 2 workouts in the last month because they are in different phases of the user lifecycle.