Everyone who has a need to calculate safety stocks knows the real version of my over simplified formula.
Safety Stock = z * √(Stuff)
This formula, rooted in the assumption of a normal distribution, has been a staple for safety stock calculation enthusiasts like me for the past two decades. However, recent trends have sparked a wave of skepticism among practitioners, prompting us to question the validity of this assumption and explore alternative methodologies. This led to various questions in my mind: Why is normal distribution so popular? And if there is an identifiable limitation to the assumption, what are some alternative ways of calculating safety stocks.
In fact, it is very easy to see why normal distribution is so popular. It is easy to use and explain. The underlying symmetry appeals to the non-statistician’s mind. Almost everyone is exposed to it from school/college days where teachers often deploy the bell-curve when assigning grades. There are easy tables available for quick calculations. However, mounting evidence suggests that underlying data may not conform to a normal distribution.
But considering proof that the underlying data is not normally distributed, is there a way to still calculate safety stocks. What if the data is very low volume as well as sporadic in timing? What if the left side of the curve is missing (for example because demand cannot be less than 0).
The answer to the above question is of course yes. One could determine the underlying distribution in the data. While statisticians may advocate for identifying the true distribution of the data, this task proves daunting for those without extensive statistical training.
For those of us who are not trained in statistics, there is a beacon of hope in percentile-based methods. It is fundamentally a data-based approach and does not require any assumption of a statistical distribution. One can look at the data and evaluate where on average is the cut-off for 95% demand (assuming a 95% customer service level). If one takes a statistician’s mindset to it and calculates this many times using many different samples (a technique known as bootstrapping), one could get to an average that is more robust compared to a single calculation. Adopting a statistical mindset, one can employ techniques like bootstrapping to derive a robust average across multiple samples, circumventing the need for distribution assumptions altogether. This approach not only yields a more reliable safety stock calculation but also opens doors to simulating stockout probabilities across lead time cycles. One can further take this calculation and run a simulation to see how many lead time cycles one would have a stock out.
Join me in my upcoming webinar as we delve into the intricacies of safety stock calculations, exploring alternative methods that break free from the shackles of the normal distribution assumption. Register here.
