For example, BeyondCore was analyzing product defects data for a Fortune 100 manufacturer. The company already knew that a certain product had a high defect rate. What it didn’t know was WHY the product had such a high defect rate. In just 4 minutes, BeyondCore revealed to them that the defect rate was high only for 3 customers and 3 countries. The product owner was in the room and when she saw the BeyondCore analysis, she immediately pointed out that the three customers BeyondCore had highlighted were mining customers and the 3 countries had really bad air quality. She explained that her product had a large fan in it and of course the low air quality in mines and these countries could easily explain the high defect rate in these cases. This insight was immediately obvious to her once BeyondCore highlighted the underlying pattern of the three customers and three countries. However, even though she had been very motivated to understand why the defect rate for her product was high, she had been unable to guess this pattern without the help of BeyondCore and her visualization tools weren’t able to help her. BeyondCore, on the other hand, looked at every possible pattern in the data without any pre-conceived notions and found the key insight, even though the expert herself did not originally guess the pattern. The key here is that, using traditional analysis, people can’t find patterns that they don’t already know.
Another example of finding patterns that are not obvious is a case where BeyondCore looked at 30 million patients in a study with McKinsey. In this case, a group of clinicians were in a room when they had an ‘Aha’ moment after reviewing some of BeyondCore’s findings. BeyondCore found that young women age 18 – 35 with diabetic ketoacidosis (DKA) had a 49% re-hospitalization rate, an astounding number one would not initially expect. Why were these young women being re-hospitalized? Because they were not regularly taking their insulin. And why were these young women not taking their insulin? As a weight loss method, because when you don’t take insulin, your body doesn’t process sugar and so you don’t gain weight. But why had clinicians not known of this re-hospitalization number nor had they thought to find it? Because usually young people have low re-hospitalization rates and women often have lower re-hospitalization rates than men. So an expert who only has time to evaluate a few dozen patterns would not think to focus on the readmission rate for young women with DKA. McKinsey’s own estimate was that they would have been able to manually test only 250 hypotheses in 4 months and so they would not have thought to examine this specific a pattern. BeyondCore, however, evaluated a million patterns without any pre-conceived notions in just 2 hours and was able to find this pattern that experts were able to confirm easily. If this data was “analyzed” the old way, this pattern would have never been discovered.
The word hypothesis comes from ancient Greek and means under (hypo) and theory (thesis). Why are we still doing hypothesis testing – the way the ancient Greeks did it? Why must we begin with a theory? Why can’t the computer look at every possible pattern and highlight the most insightful and statistically sound ones? Isn’t it time that we stop doing what the ancient Greeks were doing and rather than guessing and confirming patterns, let computers do the heavy lifting for us so that we can focus on overlaying human intuition and domain knowledge on the automatically identified patterns? Sounds like science fiction? See how business users were able to do exactly that with their unique data in just minutes in our “Your Data. Your People. Five Minutes.” challenge.