# The Importance of Understanding Process Variance

During my many years of working as senior management for food safety and quality assurance within processing facilities, the same questions always seem to come up. Why did you put my product on hold? Why are your work instructions not working? You must not be training my employees right or this wouldn’t happen? You are always telling me that this machine setting won’t work, how do you know that?

How did I know that indeed? It was a simple answer yet, very deep, and complex at the same time. I could answer those questions by understanding my process variance.

Variance in the natural world is all around us and inescapable under normal conditions. There are times when you can force and even predict a natural outcome but for most of us, things just happen. Not just happen but happen in an orderly and somewhat predictable way. Think of people being tall or short, thick or thin, the color of hair, the color of eyes. Think of trees, the same species do not grow the same branches, trunk forks, or have the same number of leaves.

Scientists and statisticians have a fancy term they like to use among many others called population variance. Populations are referred to as “Normal” or “Non-Normal” for the purpose of this discussion we will talk through variance with regard to a “Normal Population”. A Normal population can be defined visually as one whose population data fall both above and below the population mean. It is sometimes referred to as the ** Bell Curve** in which there is a probability at 99.7% of the time, that a data point measured will fall between +/- 3 σ (standard deviations) from the true population mean. Variance is a key statistical term worth spending a moment to define and then I am going to try, for the rest of the discussion, to put it in more simple terms.

Definitions of Variance for a normal population: Variance is the expectation of the squared deviation of a random variable from its population mean or sample mean. Variance is a measure of dispersion, meaning it is a measure of how far a set of numbers is spread out from their average value. (Wasserman, 2022)

Variance is used as a key component of statistics to help derive hypothesis predictions, perform process modeling, and most importantly understand dispersion within your processing system. A simple way to look at it is the square of the standard deviation for measurement within your process or σ^{2}.

So why is this important to a process? Simply, the sum of all the variance within your process becomes the true amount of dispersion from your target. These are the true constraints (or the range) of your product's predicted outcome that becomes the determining factor of your specification’s success.

It is very common and almost overlooked that we measure a product at a given step, but we don’t measure in the same way all the components that go into the total product. At each key process step, there are inputs and outputs to each of these steps each potentially influencing the overall outcome. Often when we design a process, we get a lot of information on the performance of a piece of equipment from the manufacturing, but under what conditions? We need to know the input characteristics so we can determine the input factors and maintain them to create minimally variant outputs. For example, let’s say that you have an extruding machine that forms a specific shape and size of a product. The manufacturer of the equipment states that you can run this machine at 60 cycles per minute, to that shape and size. What we do not know is how incoming temperature, the viscosity of the product, density, and coarseness of the incoming product affect the shape or cycle time as it relates to the machine's promised capability. So, running a product of one formula may perform perfectly, and changing to the next formula with different input characteristics become impossible to run, or at the very best less efficient than the original 60 cycles per minute guarantee.

Another example is to illustrate total variance issues. Let’s say you have a weight specification of 100 grams in a bag and you have designed the bag to fill to about 80% of the bag's volume. Your process flow looks like this for this example:

**Ingredient blending → shaping → baking → bagging → shipping**

So here is a typical simplified way to look at variance. The first introduction of variance is at blending. Questions to ask; are your ingredients always containing the same nutrient values (especially moisture and fat which affect viscosity and density)? Are they being blended homogenously? At the shaping step some questions might be; How many pieces do not meet shape? What is the weight range dispersion from shaping? During the baking step you might ask; Is the moisture level consistent post-bake? Is the color consistent post-bake? And finally for the bagging step; What is the true average weight per bag? What is the density of the product? What is the failure rate for bag seams? Does the product achieve an 80% full appearance?

So, you take measurements at the various steps, and you would find that each step added a certain amount of variance. You could then sum each point of individual variance and determine what was needed to satisfy the specification and make a consistent product every time.

In our example say, Ingredient blending added 2 grams of variance, shaping added 4 grams of variance, baking added 4 grams of variance and the bagging added 2 grams of variance. Adding each individual point up gives you a total process variance of 12 grams. If you remember the definition this is the dispersion around the mean target of 100 grams. In my example of a target at 100 grams, 50% of the time you would miss the 100 grams target and underfill your bag. One solution would be to correct the bagger to overfill the product by 12 grams to consistently make the net weight. This is an oversimplification being used as an example, but in this example, you only get 9 bags of product for 10 bags worth of weight. (10 bags x 100 grams = 1000 grams but if we have to overfill 9 bags x 112 grams = the same 1000 grams. So we only fill 9 bags when we should have had 10). Just one example of how variance brought a loss of profitability to a process.

What’s this worth to a company? Well, I have personally been involved with large business operations where we were able to save over $20 Million dollars a year. Typically, those companies that deploy properly determined countermeasures against variation can see savings of 20% to 30% per year.

But I indicated previously there were many ways in which variance of a process causes profitability losses. Some of those losses can be readily quantified or hard profitability losses. Some examples of these might be the following.

**Examples of quantifiable variance profitability losses**

- Throughput in pounds per hour
- Overfilling loss to make net weight compliance
- Rework costs
- Increased labor costs
- Yield loss or increased scrap
- Shipping shortages due to unobtained production schedules
- Increased maintenance downtime
- Increased raw material input usage

And then, there are some examples not easily quantifiable but can still affect the overall end profitability of a product.

**Examples of difficult to quantify or soft dollar profitability losses due to variance**

- Loss of employee morale from additional rework or overtime
- Quality complaints from end users
- Not securing repeat business from end users
- Negative social media posts
- Brand reputational damage

**Countermeasures**

Understanding your variance during the different steps in a process is key to focusing the needed resources on reducing variation and taking control of your profitability. Not every step will contribute equally to the overall variance, and you may find that there is a very simple countermeasure that can be deployed to reduce variation. While some countermeasures may be more expensive and require more time, others can be implemented quickly with very little additional cost. You may also find that some equipment in itself can be a countermeasure and reduce some of the overall variances just due to the nature of the equipment.

__Some countermeasures that facilities can almost always control:__

- Standardize the time of blend
- Temperature of mix
- Disperse ingredients evenly into the blender
- Vacuum
- Cycles per minute (sometimes fewer cycles per minute actually have better throughput)
- Employee work instruction and training
- Belt speeds
- Machine set-up
- Keep machine parts and tools sharp and be up to date on your preventative maintenance as instructed by the machine's manufacture
- Adjust machine settings per an agreed-upon schedule. Limit off-the-cuff machine setting changes

I hope that during this brief discussion I have given you pause to think about how the variation at differing points of the process can either lead to the success of a product or its failure. From my old Crosby days, he used the saying, “What gets measured gets managed.”^{1 }Bringing visibility and understanding to your process variance will allow key decision makers to make better decisions and implement decisions without fear because they will understand how that change may or may not affect another part of the process. Remember the goal here is to meet the specification every time. Hopefully, the R&D and Sales team has connected the specification with the needs of the customer in such a way as to understand what the customers believe is quality for the given price, they are willing to pay.

Where to start? I outlined some helpful tips to get you on the right path for understanding your manufacturing variation based on my many years of process improvement and troubleshooting.

**Helpful tips**

- Determine the Key performance indicators for the process (KPIs). These are the ones that without control, could impact profitably.
- Develop a strategy for taking measurements. Use an all-hands-on-deck approach. If it makes employees' jobs easier they are usually all for it. People resources are ones you get for free. Imagine 30 sets of eyes focused on a process over 1 set of eyes from a quality control technician.
- Make observations and measurements to obtain data points for those KPIs
- Sampling should be random and unbiased. If not, the data is practically worthless. It is not about getting the measurements that you want to see. It is about getting measurements for what is actually happening.
- The more data the better, but if you have to pick and choose due to resource constraints, I recommend at the very least 100 data points or observations spaced out over various times and days.
- Seasonality is most definitely a factor. Make sure you keep tracking through various seasons. I.e., facility temperatures and humidity may change season to season. Raw materials may fluctuate from season to season.
- Understand that these efforts take time. Set reasonable goals and empower an accountable person.
- Collecting the data electronically can save time. You are going to enter the data collected into some sort of spreadsheet or statistical program so why not get ahead of the game and input it directly? There are a lot of software programs available for support. I like Minitab or JMP but there are many others I’ve worked with. A strong excel user can accomplish almost all that is needed just in excel. I recommend that the reporting output from these programs is strongly assessed before selecting.
- Look for low-hanging fruit. There should be something in the data that pops out with large-scale returns.
- Control raw materials – It is being purchased, so spec it. Don’t accept ridiculous inputs with super wide ranges for key nutritional parameters. Sometimes paying 10% more can save you that 20% out that we were talking about.
- Don’t be afraid to ask for help. Sometimes even determining where or what to measure can be challenging.
- Explain results visually. Statistics are more easily understood when it is pictorially represented with appropriate titles and with full transparency; good or bad.

**Final thoughts**

There are so many additional benefits to taking this systematic approach to understand the process variance. These KPIs and measurements can become the building blocks of a quality management system (QMS) where-by decisions can be made from data and not from a “shotgun” or “kneejerk” approach. Another benefit is it counts towards passing with high marks, in any of the global auditing schemes. This data and analysis can be used to validate risk assessments, demonstrate continuous improvement, and help with answering consumer inquiries and complaints.

We understand with first-time adopters and start-up companies’ resources may be limited. Yet, typically processes variance, if not controlled adequately, can cost 20% to 30% of a company’s profitability. Good use of outside help or consulting can yield real returns. At BSM-Assurance, we have a large toolbox at our disposal, and we have tackled a wide range of processing concerns, worked to reduce variance and troubleshot various food safety and quality issues. We have the experience you need to help you limit your variance and achieve success.

**References**

1. Crosby, P. B. (1978). *Quality is Free.* McGraw-Hill Education.

2. Wasserman, L. (. (2022, June 20th). *Wikipedia*. Retrieved from Variance: https://en.wikipedia.org/wiki/Variance#:~:text=Variance%20is%20a%20measure%20of,fit%2C%20and%20Monte%20Carlo%20sampling.

**Next Time: Modeling your specifications for success or failure based on measuring your process's capability (Cpk).**

**About the author:** August Konie M.S., M.B.A has been a Food Safety and Quality Professional with various teams for over 25 years.

*Follow us on LinkedIn for the latest updates on all things happening here, at BSM Partners.*

This content is the property of BSM Partners. Reproduction or retransmission or repurposing of any portion of this content is expressly prohibited without the approval of BSM Partners and is governed by the terms and conditions explained here.