Cohort analysis has been around for a while in medicine, academic research, business, and other fields where there is a need to compare apples with apples. Its potential applications in the mobile marketing world are as obvious as they are thrilling. So what is cohort analysis and why should mobile app businesses be excited about it?
When you try to identify changes in user behaviour from tweaking or optimising your various channels and features, you'll often find that the trends or changes are too weak to draw any conclusions. This usually occurs when your user base is varied, with different levels of engagement and at different stages in a user lifespan.
Cohort analysis is designed to solve this problem. A cohort, in its strictest definition, is a group of users who share some common criteria. In a cohort analysis, you then compare these groups, observing their performance over days, weeks or months, allowing you to observe the trends and movements that are otherwise hidden.
Let's say you're a winemaker and you've just made some modification to your, er, grape-squeezing. You want to know if that was an improvement. So you pour a glass from your most recent bottle, and a glass from an older vintage and you then ask your sommelier to do a blind test. This probably would not be the most astute comparison. It's easy to see the flaw here - your vintage bottle has aged and matured, changing its flavour and you should, know as a winemaker that the flavour of the wine for your newer bottle would also change over time. However, in this moment, the bottles are simply not comparable - and with this method, they probably never will be. There is a difference between the glasses, or what we like to call ‘interference’.
Lifespans of mobile users
Your mobile users change over their lifespan in the same way. They will stay for a certain period of time, and they generate different activity when they first start using the app than after a few weeks or months. Cohort analysis is a way of removing the interference - it's as if you taste, monitor and record the flavour of your wine on each day of the during a specific period of maturation.
Let's start simple - we segment the users in your mobile app by the date on which they downloaded the app. Remaining performance data is then aggregated by the install-week segments. This already allows for a whole new definition - pick out the revenue figures from a single install week, on a given number of weeks after they installed.
Once you have these segments, you'll want to compare their lifespans. Simply line two cohorts up so that you're comparing metrics for the first week after install, second week after install, and so on.
Since you're directly comparing users at equivalent times in your app, you've removed the interference that comes out of their morphing lifespans. There are plenty of questions you can answer here - and plenty of new questions you previously couldn't ask.
Cohorts give you the ability to track user segments from a specific time period. For example, you might want to segment users based on which week they installed your app post launch, and track their behaviour as a cohort over the user life cycle. You can also test the effectiveness of different marketing campaigns to see which channels give you loyal, profitable users in the long run.
Applications?
Cohort analysis removes the confusion that can arise from running simultaneous improvements and campaigns, so you can get a better sense of what is yielding the best results, and is therefore worth investing in.
Let's say you update your checkout process, and you'd like to see if your revenues improve. As the change affects all users, we'll look at revenues across the board – with no other segmentation.
Can you simply look at your total revenues and see if there was an upward trend? No, because new users coming in would warp this, or a large group of users starting to adopt your app. Could you calculate average revenue per user? Not really, because a large amount of the users that make up this average are at different stages of use within the app. Only by employing cohort analysis can you isolate what you're looking for: Whether your update improved the way your users go through with purchases, not just if those users would have gone through with it anyway. This is a critical difference.
It's more than a lot of nifty charts - cohort analysis allows you to dive deeper and create much better segments out of your users. The only way of spotting trends in heaps of usage data is by respecting the lifetimes of those users. Only by doing that can you have a truly comparable base to optimize for growth.
By Simon Kendall, Product Manager and Head of the Analytics Operations team at Adeven.
PrivSec Conferences will bring together leading speakers and experts from privacy and security to deliver compelling content via solo presentations, panel discussions, debates, roundtables and workshops.
For more information on upcoming events, visit the website.
comments powered by Disqus