Retentionizer expects csv data to have headers, with the first column named 't' representing time periods, starting at 0, counting up. Each additional column represents a cohort to model, as a survival percentage. The first cell (at t=0) for each cohort must be 1. Survival %s are expressed as decimal numbers. You can view a valid example csv file here , which contains the two cohorts from the original Fader & Hardie paper.