I believe what they mean is that they use historical data to analyse all customer behaviour, they then apply that analysis to new customers.
So they probably rank customers in value and then look at what behaviours the most valuable had in the week, 2 weeks, month, 3 months etc after joining. They probably started with a big long list of data items, and through statistical analysis identified which items were the strongest predictors of future value, then they simply look for these behaviours in new joiners. Not an exact science but helps direct limited value add resources to be more effective.