A colleague once asked me to help him build a data model. Its purpose was to rank suppliers in a portfolio based on certain performance criteria. His scoring system was thoughtful and at first glance the ranking seemed to pass the major common sense criteria – the relative ranking reflected intuition, it highlighted known problem areas, and the highest ranked suppliers scored close to 100.
The problem was one last binary criteria. My colleague just couldn't work it out. Ideally, the model would look for this factor, mark it as yes or no, and adjust the ranking. Simply, it would split the list in two, the suppliers with 'yes' all ahead of those with 'no'. The problem was a technical question asking how to split the list into two groups, the subsets then applying the same criteria to rank the segregated suppliers.
I realized that there were clever ways to solve this challenge. You could, for example, mark the highest score for any ‘NO’, find the difference of that and a perfect score, then proportionally grade the 'YES' suppliers along a curve with those two scores as the endpoints. However, I had no time, so I just said, ‘why not add one billion points if you have a YES?’ and went back to my desk.
