Around 1964, when I was involved in starting the Bank of America’s first marketing research department I chose to work on the Bankamericard. The predecessor of the modern Visa.
I became the in-house expert on Bankamericard and was approached by an economist in the bank who was asked to do something about reducing the losses on Bankamericard. (I can’t remember his name.)
Together we used the new IBM 360 computer that the bank had which still used punch card input. We took two files of cards that included all credit application information. We put the pile of deadbeats in one category and the people with good credit payment history in a second category. (Note: the dead beat loss at the time was under 1%. That would be considered wonderful in modern credit card banking.)
We ran a linear regression on each pile of punch card credit data. The result was two equations that allowed us to create a scoring system that was highly predictive of bad credit behavior.
It was immediately implemented at Bankamericard and to my knowledge spread to the rest of the world.
As I recall there were two variables that stood out. People with bad credit seldom reported a telephone number (probably hiding from creditors) and seldom filled out the credit application in full (probably not expecting to get credit).
Many people have strong aversions to credit scoring. The hostility is probably justified. On the other hand, there were not enough people in the labor force to check all credit applications that the modern credit card generated.