Last year the bank increased fees for customers with small balances, but said it would consider each customer's entire relationship and not merely the balance in a particular account.
Machine learning distills the raw information about the behavior of each individual customer and creates associations that are accurate, but general enough that a campaign based on those associations has a good chance of success with a substantial group of people.
Under the new structure, InPhonic gets a smaller upfront fee for each handset activated, but it also gets residual payments for as long as the customer stays with the carrier.
But it is notoriously ill-suited to high-tech industries that have to tailor each item to a customer's specific needs (as in a data-switching centre or air-traffic-control system) or in which it is crucial that thousands of interrelated components all work correctly together (as in a rocket vehicle).