Bank One's PORTICO priced each cardholder's rate and credit line by Markov optimization

Bank One's card business faced a classic tension: low annual percentage rates (APRs) and high credit lines attract customers, but they squeeze profit and raise credit-loss exposure. The bank changed rates and lines with broad policies rather than account-level logic.

Researchers designed PORTICO, a portfolio control and optimization system based on Markov decision processes. For every account it uses historical purchases, payments, profitability, and delinquency risk to select APR and credit-line actions that maximize the net present value of the whole portfolio.

In competitive benchmark tests run for over a year, PORTICO outperformed the bank's existing method; the Interfaces paper reports it may increase annual profits by over $75 million. The system was built to sit inside the bank's day-to-day account management decisions.

Treat each cardholder as a Markov state

Account-level data let the model tailor price and line to each cardholder's behavior.

The Markov formulation captured how customers respond over time, not just today.

Optimizing the portfolio's net present value aligned pricing with long-term profit.

Pricing is a dynamic control problem: a Markov decision process that weighs each account's response against default risk beat the bank's one-size-fits-all rate and line changes.

PORTICO was documented in Interfaces as a model for account-level credit-card pricing, and its benchmark results showed the same data could support far larger profit. Account-level pricing and credit optimization later became standard in US card portfolios.