Suppose I were to ask you to determine a word that I am thinking of and you could only ask yes or no questions. You might start by asking, “Does the word begin with the letter A?” If I said no, you would probably ask, “Does the word begin with the letter B?” and so on. Another strategy might be to ask whether the word is an animal, mineral, or vegetable, thereby narrowing the number of possibilities.
The best strategy, however, would be to ask the question, “Does the word begin with the letter A thru L?” By asking that one question, you have eliminated half the dictionary, and you really don’t care if I say yes or no! If the answer is yes, the next question would be “Does the word begin with the letter A thru F?” and so on.
This same strategy can be used in determining the root cause of a problem. By eliminating what it’s not, you converge on what it is. Understanding how to do this requires you to think about the various families of variation (FOV) that exist in your process. Let me explain.
In any process, there are many inputs that could have a large effect on the output. We separate the set of all such inputs into two or more families with common features such as the location or time frame in which they act. Then we use data, investigation, and knowledge of the process to rule out all but one family where the root cause lives.
Generally, we think about FOV starting from micro to macro. For instance if we were turning steel rods on a lathe, the first FOV would be within piece or within rod variation. We might measure each end and a point in the center. The next families would be piece-to-piece variation, hour-to-hour variation, shift-to-shift, day-to-day and so on.
We use this same idea to search for the root cause. We divide the set of all causes into families and then conduct an investigation to rule out all but one family. We repeat this exercise on the remaining family until the root cause is all that remains.
Using the method of elimination, we do not start by listing possible causes and classifying them into families. Instead, we start by considering broad categories that define the families. As we obtain more process knowledge, we subdivide the remaining families, using a diagnostic tree, to further narrow down the suspects for the root cause.
In my experience, this convergent process is extremely efficient and results in finding the root cause much faster than many of the other conventional problem solving methods.