≡ Menu

Premier Bankcard and building a good customer score #paw


Rex Pruitt discussed Premier Bankcard’s development of a good customer score. I have written about how Premier Bankcard is putting predictive analytics to work as I used them in my presentation on putting analytics to work earlier. Job 1 was to define a good customer – find the definitions that the CEO, the CFO, heads of operations had and blend them into something that would work for everyone. Given Premier Bankcard’s business model this is unique to them and is based on a set of characteristics. The key components were revenue (how do we make money from them), risk, behavior (what do they do to us/for us), loyalty and were combined to measure how good a customer someone is. Using this score they can rank order customers from best to worst. Developing the model involved a variety of SAS tools.

Once the model was developed it was essential to show that the new model would improve business decision making so they used a comparison of third party scores and existing internal scores to see which ones predict their best customers most accurately. The new model did a better job of predicting the best customers but nevertheless was a big project as all the operational systems were already optimized for the third party and existing scores.

To show why the score would help he ran some analysis to show the difference if they used the model. Churn is a big problem for Premier as many of their customers are sub-prime and "graduate" to other products. He showed that it would reduce the churn of "top good customers" after 2 years by just 10% and increase revenue by $15M. 25 days of IT work was needed to drive this into the systems.

The second example was an existing cross-sell program. This program found people who might be eligible for a second card. The data showed that if driven by the new score it would increase qualified credit offers by 2% over the program driven by old scores. Being able to target these extra customers they could generate $2.3M in additional revenue. If they used the score across the whole portfolio they would be able to generate $24M in additional revenue because the old program was issuing second cards to people who weren’t using them profitably – cards that were making a loss.

Even though the new score was better there was still a great deal of management support needed and a great deal of IT change. Senior management time is really hard to get but eventually the CEO saw the potential and asked why it could not be tested? This CEO support was essential for driving testing and ultimately adoption. Despite implementation challenges and delays, the program is showing great returns.


Comments on this entry are closed.