The Demand Driven Supply Chain. Part 2: Variability and Uncertainty

In Part 1 of this multi-part article, I took an information centric view of the supply chain. And said that demand is not about forecasts or orders. Demand is, at its heart, a signal. Being demand driven starts with identifying the best possible demand signal for every node in the chain.

In this part 2, I want to dive deeper into the characteristics of the demand signal – specifically about the difference between uncertainty and variability. When discussing supply chain topics, it is routine to get these two measures mixed up. To be fair, they often go together, so it’s not entirely surprising. But if we want to be demand-driven it is important to tease out the subtle but crucial difference between the two.

Variability and Uncertainty Are Not the Same

Imagine if your data looked like this (a sine wave). Is it variable? Yes. Is it uncertain? Not at all.

Now imagine if your demand history looked like this (white noise). Is it highly uncertain? Absolutely, by definition. Is it variable? It depends – how far the noise is from the mean.

Now take an example of this pattern below. It is clearly variable and noisy.

Of course, real world data is never as clean-cut as the above examples. Consider the graph below. It looks pretty random at first. But then you see that this is a graph of US vehicle sales by year. And the steep drop offs correspond very closely with recession years.

 Understanding which part of your data is due to variability and which part should be attributed to uncertainty is crucial. The difference is that variability is explainable (and by extension predictable) - by driving factors (a prime example being seasonality). Uncertainty is, by definition, the portion of your data that is not explainable (by known factors).

?The First Deadly Sin

A real-world example of the confusion between variability and uncertainty can be seen in ABC-XYZ classification. This commonly used method groups items according to importance (ABC class) and also according to predictability (XYZ class). The idea being that demand planners should focus their attention on SKU’s with high value and low predictability. SKUs with high predictability should be left to statistical forecasting methods.

Sounds entirely reasonable, till you observe that the most common metric used to measure predictability is the coefficient of variation (CV). This is odd because the coefficient of variation is not a measure of predictability at all, it is a simply a measure of the ‘spread’ of the data.

This becomes obvious in the extreme – Consider the sales of Stovetop Turkey Stuffing (a Kraft Foods brand) in the US. I would hazard a guess that about 90% of its annual sales is concentrated in the four weeks before Thanksgiving. Extremely variable on a month to month basis. But uncertain? Not so much.

I think I’ve made my point. Variability is easy to measure, and it is can be an important metric in itself. But it does not tell you how uncertain a dataset is.

Misidentifying variability and uncertainty is the first deadly sin of the supply chain. The result is a poor demand signal and a cascading effect downstream including too little or too much buffers, poor customer service and high operating costs.

The God Particle

Which brings us to the next logical question. How then do you measure uncertainty? 

Take a look at the graph below. Don’t worry too much about the strange symbols and numbers, the main thing to notice in this graph is that tiny bump in the black dots (observed data). The red dotted line indicates the signal that was expected to be seen from all the previously known processes (fitted model).

Normally one would be tempted to ignore that tiny bump as a statistical fluke. Noise as it is. And remove the ‘outlier’ from the data to get a perfect fit! But if you did that you would have missed one of the most important discoveries of the 21st century.

The graph is a summary of 2500 trillion particle collisions at CERN that led to the discovery of the Higgs Boson (a.k.a., the God particle) and proved the theories proposed Peter Higgs and Francois Englert nearly 50 years earlier, and for which they were awarded the 2013 Nobel Prize for Physics. That little bump in the graph is due to the Higgs boson, essentially the reason why objects have mass. So yes, kind of important!

 The point of this example was to show that what is ‘noise’ (uncertainty), and what is a ‘signal’ (including variability) is contextual. The bump was noise only till the time it was explained by a causal variable.

In summary - Only when you have an underlying model with assumptions of drivers and its effects can you distinguish between the variability and uncertainty. Uncertainty is what’s left after you identified the signal. It’s a residual.

Or, as the epigraph of the 2013 Hollywood thriller ‘Enemy’ read – "Chaos is order yet undeciphered”.

The Serenity Prayer

Giving up on forecasting because of VUCA (Volatility, Uncertainty, Complexity & Ambiguity) is a cop-out. At the same time, ignoring the real effects of uncertainty is no solution either. You now have access to virtually unlimited data and advanced tools that previous generations of supply chain analysts could only dream of. Use these tools wisely.

Nobody said being demand driven was going to be easy. Which is why supply chain planners need their own version of the serenity prayer - “God, give us the courage to predict the things we can, the serenity to accept the things we cannot, and the wisdom to know the difference.”

#DemandDriven #SupplyChain #Variability #Uncertainty #DeadlySins #VUCA #DriverBased #Forecasting

?Disclaimer: This publication does not represent the thoughts or opinions of my employer. It is solely based on my personal views and as such, should not be a substitute for professional advice.