How to calculate retention and churn uplift
How to calculate retention and churn uplift
There is a good posts around the web on user retention for websites but Mobile App developers initially focussed for a (long, long) time on “Downloads” as a measure of credibility. Now the App industry is maturing rapidly to understand that every App is a leaky bucket – that you lose users over time and you need to proactively plug the gaps by engaging your users and adding value to their day.
Meet the Churn Curve
In general a graphical representation of users leaving your app looks like this:
Its a good chart for communicating the concept but its over-simplified – why?
- That chart was like you took a snapshot of only the users on September 1st and then how long (in days or weeks or months) they continued to open your App.
- You are actually acquiring downloads and users (hopefully) everyday
- Depending on your marketing activities and word-of-mouth – different users will have different retention characteristics
- The utility of your App. A game might have an average lifecycle in days – a public transport App is retained as long as the user commutes!
- lots of other variables like holidays, seasons, relevance of your App to time of year (e.g a Football League App’s cycle is very different to a Black Friday Shopping App’s)
Getting more specific with Cohort Chart
So continuous acquisition of users means that a cohort chart is more relevant.
So the cool thing about this is you can see the churn rate and/or retention rate for users that downloaded, 12 weeks ago, 11 weeks ago, 3 weeks ago etc etc.
Just for clarity:
(R)etention = 1 – (C)hurn
if you have 5% churn rate per week, then your monthly retention would be roughly:
100% – (5% * 52 weeks)/12 months = 78.33%
This is a more concise way of describing the success rate of your “Engagement Strategy” than purely talking about MAUs (Monthly Active Users). Don’t get me wrong MAUs is a much better metric than the vanity metric of “Downloads”. People love to talk about “We’ve got over 300,000 downloads” but don’t know or won’t say what the MAUs are. It reminds me of the joke….
“A million people walk into a Silicon Valley bar. Nobody buys anything.
The bar is declared a huge success.”
The Cohort Chart can be represented by the following graph:
Its an interesting chart but not very actionable – here is an average across the datapoints.
You can see we are back with a simple understandable curve similar to our original chart which shows you can either express it as a “snapshot” or an average across the dynamic user-base. As you can see its easily shows that around Week 6 we would want to understand why users are leaving the App. We would revisit our engagement strategy to “flatten the churn curve”.
Getting a baseline of Retention or Churn
You might have missed before how I glibly provided an example of “5% churn rate”. You can see from this chart that churn varies from week-to-week and for cohort-to-cohort. So its not easy to just say “5%”.
Today, we think the best way to define (R)etention or (C)hurn is to run a “line of best fit” through the points you are interested in. This means you could:
- pick one cohort (e.g Week 12)
- pick 4 consecutive weeks of cohorts
- pick the average chart above.
Line of Best Fit (Linear Least Squares)
The simplest and most common method of getting the line of best fit is not actually with a ruler – but using a calculation called “Linear Least Squares“. You can get complex with this stuff (non-linear etc) but for the average developer, product manager, marketer of an App, they just need to get their baseline and start figuring out how to improve it (get uplift!). Picking the average chart above (12 weeks of cohort data) we get a new chart.
Whats most important to App developers, product managers or marketers is the slope/gradient of the line. Your mission….should you choose to accept it….is to reduce the gradient of the line. A gradient of -3.5 is much better than a slope of -5!
Weekly Churn and Retention
Because each cohort starts with 100% installs, the slope is actually the weekly churn rate so a slope of 5% then C=5%. As discussed before then R(monthly)=78.33%.
Calculate Retention Uplift (RU)
Phew!!! thats a lot of foundation work to get to a very simple uplift calculation – as I said before:
“Your mission….should you choose to accept it….is to reduce the gradient of the line”
You can take any two equal samples of your C or R to get the uplift:
Choosing the right period
I’m a big believer in looking at short and mid-term periods. In the example above the 12-week data is very useful because we see the big picture. We discover the drop-off around week 6.
Further, in this example if we only look at the first week we don’t learn much because retention is generally 100%. (you don’t yet know if your users have left you).
However, its important to not wait for 12 weeks of data to make decisions. I think you should be looking at 4 week periods as well to get a sharper period.
For calculating uplift you can choose any two periods (even overlapping) but obviously your “later” period, should be including this week – you want to know how you are tracking now!!!
Measuring “Active Users” or Uninstalls
The big difference for Apps is that you might only want to engage users a few times a month. I’ve talked before about how stupid it is to try to keep people engaged by spamming them – you are not Facebook so don’t be pushy about your relationship with users.
For this reason, there is two ways to measure user engagement:
- Have they opened the App in that week?
- Do they still have your App installed?
StreetHawk allows both to be tracked and in fact (surprise!) the chart above is actually tracking Uninstalls – a very powerful metric for just how MUCH users don’t love your App anymore 🙂
You can always expect “Active User” Churn to be more brutal than Uninstall Churn.
StreetHawk makes this easy
I hope the post has been useful – whilst the content might be a little detailed, you don’t really need to think about it. The StreetHawk platform does all the leg work in helping you understand and optimize to “flatten out the churn curve”