I've been heavily involved in building marketing, financial, and other mathematical predictive models.

In general, what customers spend (revenue) over some unit time (week, quarter, year, etc) is more important than how many customers you have, and is easy to define and measure. Where this question often comes up is in calculation Customer Lifetime Value (CLV) which can be used to justify greater marketing spend, even spending at a short-term loss, if it is believed customers will remain with the firm over long periods of time.

One of the ways customers are tracking is using the concept of a "Cohort" which tracks which point in time that group of clients first became customers. Similarly, you need a definition of when a customer is no longer a customer. (Perhaps they don't renew for a period of time, no longer visit the website, etc. In your case, not using the smartphone app for a given period time might be exiting a cohort.) You can then track customers in a given cohort. (The numbers will always decline.) If the customer renews after exiting the cohort, they become a customer in a new cohort. Your total number of customers at any given time is the sum of customers across all cohorts.

In your case, you may have multiple groups of cohorts. One cohort tracks air might track app installations. Another might track app usage, with customers entering and exiting cohorts after renewing their use following a lengthy absence.

This is a common way to build models of customer behavior. It may be far too detailed for your application. However, by constructing the dataset this way, a skilled modeler will be able to slice and dice the data to their needs, and thereby come up with estimates that are useful in predictive modeling.

I'd be happy to discuss this further in a Clarity call.