Persistent decay curve patterns

If you look at the number of users that drop-off from one revenue event to the next month over month, you have a very reliable means for predicting how much value the entire cohort will be worth.

When you're working on a timescale of years, cohorts tend to be inflexible, especially if product changes aren't really affecting the users' path to value. You find that each weekly cohort decays at roughly the same rate (a similar percentage of people drop off at each revenue event). This makes them great for estimations.

The amount of people who you retain in month six from a cohort from three weeks ago is very likely to be similar to the amount of people that you retain in month six from a cohort from eight months ago, percentage-wise (even if your acquisition that improved along the way).