The fancy tools, especially the ones that mathematically model process behavior, can’t solve all of the problems we run into in our organizations, especially when we step away from specific manufacturing processes and into more people-process types of problems. However, the mathematical tools are based on fundamental problem-solving methods. When the tool won’t work, look to the method the tool uses.
It happens to me almost everywhere I visit. I’ll be presented with a complex process problem, but the nature of the problem won’t lend itself to the tool I would love to be able to use to solve it. For example, often, I would love to use multiple linear regression techniques to determine the root causes or primary influencing factors of a problem, but the nature of the problem doesn’t lend itself to the type of data a regression model needs. So I can’t use regression mathematics, but I can still emulate the regression method to sort through the factors to find a root cause to address.
I’ll walk through an example so that my statement can be understood. Pretend that you are a process problem solver for a manufacturing business and you are called to the production floor by the assembly team to address a problem. A plastic component that is part of the product assembly is demonstrating some unacceptable variation.
The component in question is a coated or painted plastic part. Some of the parts look great, but others are too pale or the coating is coming off in assembly. Something is wrong with the finish. A discussion with the component process experts reveals the following list of potential causes for the part coating variation problem.
- One or more suppliers are doing things differently
- The problem may be related to specific batches from specific suppliers
- The problem may be related to specific machines from specific suppliers
- The problem may be related to composite raw material makeup
- The raw material (supplied to the molder service supplier) may come from multiple sources
- The raw material may come from multiple machines
- The variation could be between different batches of raw material
- The parts may not have been cleaned properly before coating
- The coating chemicals or dies or mix might be out of control
- The coating may not have been cured properly
- The specification for the parts might not be clear
That’s more than enough factors for the sake of discussion. If this were a simple chemical or machine process, it might be easy to identify several input factors and one or two output measures, collect some data and arrive at a nice multiple regression model that tells us the relative influence of each factor on the output. Unfortunately, we don’t have a simple, controlled process; we have a list of factors that spans everything from our own specifications to a supply chain of multiple raw material and manufacturing suppliers.
It’s not simple, but we can still use the multiple regression approach to quickly dive into root cause. Ultimately, what does the multiple regression tool do? It examines how differences in each factor correlate to differences in the output. Those factors that match up to changes in the output are identified as the primary influences on results. We can do that without the statistical math.
We have a list of factors. Let’s start looking for the connections. First, let’s do what we would with a regression model and look for quick and easy ways to simplify the model by eliminating unnecessary factors. In this case, perhaps the easiest thing to do is look at the specification. We have it on file and within an hour we can probably determine if it might be part of the issue.
Let’s say that we examine the specification and, while it might be tightened up or clarified a little, it doesn’t appear to be driving the problem. We set a task for an appropriate person to improve the specification and return to our list of factors. Let’s simplify some more.
The best advice I have to offer, in a written post format, for simplifying the list of factors is to compare our problem with a game of “20 questions.” If your colleague picks a number between 1 and 100 and your job is to identify that number with the fewest yes/no questions asked, how would you start? I like to ask if it is even or odd. Others might choose to ask if it is less than 50. Either way, the question effectively splits the inference space in half. This is what we want to do with our plastic part problem.
Perhaps the first, easiest thing to do after eliminating our specification is to see if the problem parts can be narrowed down to a specific supplier. Let’s say that we are not so lucky. There are four suppliers and the problem shows up with two of them. The fact that two different suppliers are sharing the same problem might suggest that the problem is a common input to each. Let’s look at raw material sources.
This hypothetical thought exercise could go on for pages. Maybe the two suppliers get raw material from at least one same source. Or, perhaps they don’t but they both have a similar process control problem. The key to answering the questions efficiently and eliminating factors until the truth is revealed is to look for the most efficient way to reduce the inference space or eliminate factors.
Generally, start with those things that are most likely to be common across the inference space. In the case of numbers, we try to split the space in half, then in half again, then again, until we get to the right number. In the case of our plastic part, we look first toward the specification because it affects everyone and it is also within easy reach to investigate.
Then, we look to the next level of inputs, the part suppliers, because differences in processes or machines or raw material will be determined by that factor. If we can reduce the number of suppliers that are involved, we can eliminate how many raw material suppliers or machines or processes we need to examine to find the root cause.
For those who are scrutinizing my advice, this approach is not only how regression works at a mathematical level, it is also how the Components of Variation method works at a combined mathematical and logical level. That’s my point. We have and learn some fancy mathematical tools. Those tools are simply built upon fundamental problem-solving methods. The method, not necessarily the math, is how the problem is solved. The math simply allows us to explore numerical data using problem-solving fundamentals.
Therefore, we can reverse engineer, if necessary, the method our favorite tools apply as a form of inspiration for how to solve a problem that doesn’t include numerical data. I know that sounds like a long way around a discussion of problem solving methodology, but I perceive that I’m not the only one who says to himself, “If only this were a numerical data problem I’d know just how to solve it.”
If you know how the statistics would solve it, then you can figure out how logic and reason can also solve it. Don’t be limited by your toolbox. If you have a problem your tools won’t solve, take a step back and use the method if not the tool. After all, the method came first.
Our usefulness as process improvement or business performance improvement experts is determined more by our ability to solve problems than by the diversity of our tools. Method is more important that tools.
Stay wise, friends.
If you like what you just read, find more of Alan’s thoughts at www.bizwizwithin.com
Filed Under: Rapid prototyping
