Newsletter 02/2018 - Statistics in Decision Making, Article 2

2frame Analytics' newsletters are back!

This is the third in a series of several articles on Statistics in Decision Making

In this second article, we will talk about this precisely: the company's data as a strategic asset. And to talk about that, let's use the story of a famous US credit card company to help.

Until the early 90s, the U.S. credit card companies used a uniform pricing policy for their products for two main reasons: they did not have adequate information systems to handle different pricings on a massive scale, and the area managers believed that customers would not accept price differentiation.

By 1990, when information technology was able to work with more sophisticated predictive modeling, Fairbanks and Morris proposed to various financial institutions other types of products (diversified pricing, credit limits, balance transfer from another card to lower interest rates, cash back, loyalty programs, among other things). None of them accepted the proposal and the two were called at Signet Bank after all (a small Virginia regional bank) to put into practice their ideas.

Signet did not have the data needed to implement the new strategy, since the data it had all came from products with equal pricing (it was not just Signet specifically, no bank at all had these data at the time). The consultants implemented experiments to collect data under a sample design conceived by them and that was necessary to ensure the data acquisition, in order to be able to implement the strategy.

Signet endured a number of years of default increase (which doubled during the period) and even dealt with the dissatisfaction of shareholders and stakeholders, but once the database was "acquired" and the new modeling was implemented, everything was reversed. In fact, it was so successful that Signet's credit card operation was separated from the bank's operation and became Capital One, one of the largest US credit card issuers. And they still perform tests with experimental design: for example, in 2000 they performed 45,000 tests on their clients' base.

And what data did they want to acquire that proved to be valuable for the strategy's deployment?

In the case of Signet, they had to acquire use behavior data of different types of credit cards among diverse types of clients. The use of sociodemographic data is very important for the model to be able to separate the client groups, but the individualized (and randomized) behavioral data greatly increased this separation's performance.

What about in your company, what data is stored and can be used for decision making?

Read more about the subject of this Newsletter:

1) Data Science for Business, Provost & Fawcett, 2013

Originally posted on 02/16/2018