You're evaluating a product change when a clean randomized test was not feasible. Usage shifted over time for both exposed and unexposed users, and you need a way to separate the product effect from the background trend.
How would you use difference-in-differences in a product growth setting?