I need to cluster new customer according to their future potential, but I have only information about their first transaction.
I have access to all transactions for the other customers.
So I can do a non-supervised learning on the customer who subscribe since at least 12 months in order to group them according to their potential. Then do a supervised learning to recover these groups, but only with variables from their first purchase.
But this kind of method can lead to a conceptual drift because a lot of new customers appear every day in the databases
Do you have any other idea?