Analyzing Cohorts, Funnels and Event trends are intrinsic to any MA solution meant for consumer businesses.
They are the fundamental capability so they individually don’t stand out as a differentiator. The key leverage is in the granularities and the nuances that the platform is providing on top of these key features.
Apart from the aforementioned ones, there are several other analytical methods like audience segmentation, Behavior, Flows etc. We would discuss some in upcoming chapters where they fit more appropriately.
In this chapter we are going to evaluate Cohort, Funnel and Event trends under the following three headers:
This chapters covers following topics
1. Event trends:
2. Funnel analytics:
3. Cohort Analysis:
1. Event trends
What is it?
Event trends or analytics is the study of the events performed by users on your app and engagement channels. Or just any medium where you are able to capture user actions in the form of events.
You could also call it action analytics too, as ‘events’ are anyway nothing but just actions. These action signals help us understand the behavior of the user on your app and where the work is required.
- Breakdown of app installed by countries (analyzing events across single dimensions)
- Breakdown of app install by countries and further by devices (analyzing events across two dimensions)
- What are the top performing events weekly, monthly, daily etc
- Split of critical events such as “checkout” across OSes.
- Distribution of app crash across device types
- Average number of products added to the cart daily/weekly/monthly. For instance, the following chart from Amplitude’s demo account tells the the average number of songs played daily.
- Can you analyze events across two dimensions?
- Can you add filters to users while analyzing the events?
- Do you allow adding annotations to the chart?
If you go through the use-cases again you would observe that in the #2 we are analyzing the same event that we did in #1 but only this time against two dimensions.
We would frequently require analyzing events across multiple dimension and having the flexibility to do so is must.
Let’s narrow down the said use-case further- analyze breakdown of account creation over countries split by referral campaigns only for users who have subscribed to the newsletter.
Useful, not a deal breaker but quite helpful for large teams.
2. Funnel analytics
What is it?
Funnel means different things elsewhere. But in marketing, it is the retroactive analysis of the set of events arranged in a certain order.
It essentially lets you visually track how users are dropping out along the way towards conversion, thereby giving you an idea of the bottlenecks that need to be fixed.
For instance, from the above funnel, we can conclude that stage 3 needs some rework. It also says that we are successfully able to generate traction but not able to convert into $ because the bottom of the funnel is depressing.
It is majorly used to determine drop-out rates. The following use-cases would bring some clarity:
- How to optimize the checkout flow?
- What is the average time taken by users to complete all the steps toward conversion?
You would create a funnel to make your conclusions. Consider the above hypothetical funnel
Email clicked (100%) -> Add to cart(26.18%)->Order complete(3.60%)
We are seeing that a significant chunk of users are dropping off at the ‘Order complete’ stage clearly suggesting the problem area which needs optimization.
Funnel analytics gives us the average conversion time. Say, you are in VOD business like Netflix and you wish to know the average time taken by users to watch video post install.
Key questions/ Considerations
- Do you provide comparison in funnels at the visualization level?
- How do you account the users whose flow is bit out of order.
- Do you show the median of ‘time to convert’
- Do you show the average and likewise median for each step in the funnel
You would want to compare multiple segments against a common funnel.
Say for the funnel used in the example above, you would wish to figure how the drop-offs change for “returning users” vs “first-time users” or ‘paid search’ vs ‘organic search’ acquisition. The side by side visualization of the two funnels gives an understanding of how the two segments are performing and where we need to put in some work.
As in, instead of A->B->C->D, suppose the user traversed through A->B->C->A->D. The flow is not strictly linear according to the definition.
Would he still be counted in the funnel? In case, you count only those users who have strictly moved through the defined flow then do you provide an option that enable us to accommodate the latter category users too?
Average time to ‘convert is misleading’. It is tempered by users who have too high or too less average time to convert. Observe the following funnel for clarity:
As per the above funnel of a VOD account, the average time for the user for playing the video post app launch is approximately 9 hours. The number is this high because we have taken the time-to-convert to be 7 days and there will always be outliers who would play the video close to 7 days post app launch. This would inflate the average thereby making it imperative to observe median while doing funnel analysis.
3. Cohort Analysis
What is it?
Cohort Analysis is the cornerstone of retention analytics. It helps us analyze the behavior of a certain group of users over time.
How? Cohort is a group of users did an event within the defined time period. So cohort analysis gives how frequently users in a cohort are performing certain action over their lifecycle.
At its most basic it tells you how well your users are sticking with your product over the span of, say, 30 days. The inferences you hear that X% of users uninstall the app after Y days is the product of Cohort Analysis. Follow the use-cases below to understand the nuances right.
Suggested read- Easiest Guide to Cohort Analysis (With Popular Use-Cases)
In a typical analytics tool, a cohort chart is created by specifying the first_event and return_event. First_event defines the cohort, say cohort of users who made ‘1st order’ on day 0.
Return_event is the action that you wish to track over the cohort’s lifecycle, say 2nd order. In the following chart, we are analyzing how frequently is the second order being performed by users who made first order, over the span of 7 days (the defined time period).
This provides us infinite ways to analyze retention over time which explains why it is called the cornerstone of retention analytics.
- Retention trend of users split by acquisition campaign
- Retention pattern of new users from different OS
- Analyze the Android cohort in the above to see the day wise trend of all the Android users
- Can you compare charts for multiple ‘return events’ on a single view?
You basically wish to analyze the frequency of two actions for a common cohort. Say you wish to compare usage of COD and Credit card for the transaction by new users over last 7 days for ‘new users’.
Cohort analysis is again a common feature but like we have discussed before- granularity is the key.
- Can we add filter to the user while creating the chart?
- Can we add filters to the ‘first’ and ‘return’ events
- Can we compare two events on the same chart?
Ability to add filters that we discussed in funnels is highly relevant here as well.
Refer to the 1st-2nd reorder chart again. Suppose, you wish to analyze the reorder rate for only those users who were acquired by Facebook. Makes sense to analyze its ROI.
To do that, open the filter and specify the condition for the referral attribute.
Important feature, isn’t it?
Say you are checking the occurrence of ‘checkout’ events among ‘new users’.
Now, you may also find it useful to compare the ‘added to cart’ event against the same set of users or ‘paid via COD’ event or anything. Basically comparing multiple events against the same cohort has a strong business use-case. Check if your analytics dashboard allows you to do that or not.