Measuring and optimizing user retention, churn, and average revenue per user (ARPU) is of vital importance to the success of a web application. Not infrequently the cost to acquire a user is above and beyond the value or revenue that you get from that user – and you can’t make that discrepancy up on volume. If you bring in a user and they come back a second, or a third, or a hundredth time (I’ve certainly been to Amazon and Facebook more than 100 times) – you can amortize the cost of acquiring that user down to just about nothing.

Your business has a steady-state user base that starts with new user acquisition, which is dampened by loss in your initial conversion funnel, and balanced against the churn-out of existing users

This is just a long-winded way of saying that stickiness really, really matters, and you need to iterate your product to improve it, for example:

This graph compares three sample user retention curves over a four-week period after the week of account creation. Lets say the starting point for your web application is the blue graph, where 40% of users who join in one week return during the second week, 20% during the third week, 10% in the fourth week and 5% in week five.

Then you make some changes to your application, your user retention curve now looks like the red line, 50% of people come back in the second week, but the attrition and churn in subsequent weeks is faster than you had before. Perhaps you bought traffic and those new people were outside your core demographic, or perhaps you started sending weekly emails to users that brought them back a second time, but that started to rub people the wrong way.  The total percentage of users returning over that four-week period is higher – the sum of the percentages for the period is 87 versus 77 for the blue curve. But there’s a concern that as time goes on, every user is going to get sick of your application and leave for good.

Then you decide to tone down the aggressive emailing, you focused your messaging down to a narrower but more passionate audience, and added some really valuable content that you kept refreshing every week. Now your user retention curve looks like the green line. Only 30% of your new users come back in the second week, but people fanatically keep coming back again and again. The sum of the returning user percentages for the green line is 82, less than the red curve, but more than the blue curve. But looking forward, you can clearly see that as additional weeks pass – your active user-base is going to keep building, and the sum of percentages for the green curve will quickly outstrip what you would have had if you were living on either of the two previous curves. For the mathematically minded among you, that sum of percentages can be thought of as the area under the curve, or the integral of the user retention curve – there, you just did calculus!

Since blist is a collaborative product, we get a significant number of new users from existing users sharing blists and inviting friends and colleagues. And a significant amount of sharing and inviting happens after users spend a few weeks exploring blist, familiarizing themselves, finding value in the single-user case. If our users don’t stick around for at least weeks, we miss out on additional new users.

It’s well understood that breaking down your conversion funnel into steps and measuring the drop-off rate at every step – and removing friction at every point – is critical to succeeding on the web, and it’s equally important to measure and iterate on your product for retention as well, minimizing churn, amortizing the cost of user acquisition, maximizing the lifetime value of a user – and finally your average revenue per user or ARPU.

The right way to measure changes in user retention over time is ‘cohort analysis’ – comparing the percentage of users who, say, created accounts in one day, or week, or month who log-in, or take some other action, like buying a product, in subsequent days, weeks or months with other users who created their accounts in earlier and later periods. A good time-scale to start with for most web applications is week over week.

The basic question to ask and answer is: “Is the percentage of new users from last week who came back this week higher than the percentage of new users from two weeks ago who came back last week?”

For further reading, Josh Kopelman and Andrew Chen and Dave McClure have previously contributed great resources on retention, cohort analysis and ARPU.