Why Variability Matters

Real processes do not run at a steady average. Arrivals cluster, some tasks take longer than others, and resources are not always available. That variation is what causes queues, delays, and lost capacity, and it is the part a spreadsheet cannot capture.

Why averages mislead

Plan with averages and the math looks fine on paper. But the moment items arrive faster than a step can handle them, a line forms and the wait grows. When items arrive slower, that idle capacity is gone for good; you never get it back. The higher the variation, the further an average-based estimate drifts from reality, often by a wide margin.

How ProcessModel handles it

ProcessModel models variation directly instead of averaging it away:

• Distributions describe how a time or quantity varies, rather than a single number. The Expression Builder helps you fit one to real data or build one by hand.
• Replications run the model several times with different random streams, so you see the range of outcomes and how much confidence to place in them.

Model the variation, and the results start to match what actually happens, which is what makes the predictions worth trusting.

LegacyHow this worked in the previous version

The Problem

You might notice the spreadsheet calculations for production are always wrong. The higher the variability in the system, the lower the accuracy. What’s worst is the spreadsheet always says production should be much higher. Some of the calculations are off by more than 50%. Predictions of proposed system performance aren’t even close. Why?

Most people generally don’t know what to do with variability, ignoring it. Instead, they take the average arrival quantities, processing times, and resource availability and use the calculation to arrive at a prediction. Unfortunately, these numbers have been so wrong that management has lost confidence in the projections.

Let’s illustrate the problem with two distributions (formulas accounting for time variation) next to each other. The most straightforward system like this is a single queue feeding a single server. People arrive at the line according to a variable rate and process at the server at a variable rate.

If people arrive faster than the server can process, they wait in the line - their production time extends. However, production capability is lost if people enter the line slower than the server handles each customer. That production capability never recovers - it’s lost forever.

Open the “variability kills” model and simulate using averages. Then change the model to include the distributions shown on the layout. Answer the following questions.

Question	Average with Variability
What is the expected production for 40 hours?	______________
What is the maximum number of people in line?	______________
What is the longest wait time expected?	______________

There is more to discuss regarding variability, including how to create distributions, replications, and understanding the output report.

This example shows a problem with only two distributions in the simulation. Can you imagine the potential variability problems from arrivals, resource availability, processing times, and different types of items working through an extensive process?

Variability is crucial to understanding the behavior of a system and in developing models that predict production capability.