How do I measure the performance of a mobile app?

Retention is the honest headline figure. Measure the share of people still using the app one day after installing, then at seven days, then at thirty. The shape of that curve tells you whether the app does something worth returning to, and the drop between day one and day seven is where most apps reveal that it does not. A high install count with poor day seven retention is not a marketing success awaiting optimisation, it is a product that people tried and abandoned.

Then measure whatever the app exists to produce. If it is for booking, the number that matters is bookings made in the app. If it is for reordering, it is repeat orders. Sessions and time spent are ambiguous, because more time can mean engagement or can mean somebody unable to find what they wanted. Tie the measurement to the outcome rather than to activity.

Watch the technical numbers too, because they cause the abandonment that the retention curve records. Crash rate, how long the app takes to become usable after launching, and where in a multi step flow people stop. A crash rate above a small fraction of sessions will generate poor reviews faster than anything you do to the features, and app store ratings feed directly back into whether anybody else installs it.

The uncomfortable measure worth taking is comparative. Look at what the same customers do on your website. If the app is not producing materially better conversion, retention, or order value than the mobile site, it is an expense rather than a channel, and knowing that early saves considerably more than any improvement to the app itself.

Instrument the app before launch rather than afterward, because retrospective measurement is not possible. Analytics added in month three tells you nothing about the people who installed in month one and left, which is exactly the group whose behaviour would have been most informative. Decide what you need to know, add the tracking for it before the first release, and accept that you will want something you did not think of. The cost of over instrumenting slightly is far lower than the cost of a blind first quarter.