Compute the days between neighboring purchases for each identity; for example, drop same-day duplicates and see how often they return.