The Innovator's Solution


Why You Should Embrace Uncertainty in Demand Forecasting

Posted by Joe Shamir on Nov 1, 2016 4:55:00 PM

Embrace the wisdom of uncertainty.png It's human nature to look for certainty. We believe that if we can predict the future with perfect certainty, then we will react and prepare accordingly. As such, we tend to put all our planning effort into making better predictions. The problem is that after we invest considerable effort predicting one aspect of a future event, lots of uncertainty remains.

Let's take the simple problem of predicting the outcome of a football game. By investing a small amount of initial effort, we dramatically increase our chances of predicting which team will win. But diminishing returns rapidly take over. No matter how much more effort we put in, the chance of accurately predicting the exact score becomes very small. That is, there is a lot of uncertainty that simply can't be modeled and therefore accurately predicted.

This explains also why supply chain planners struggle to improve their forecasts and end up hitting a ceiling. Uncertainty prevails in supply chain – much more so than in football. This is due to the inherent volatility and randomness of thousands, or even millions, of individual buying decisions and supplier activities. Lumpy “long tail” demand is exacerbated by rapidly changing consumer tastes and demand shaping through our own and through our competitor’s promotional activities.

For instance, our dairy customer Granarolo reports that promotions for a specific product (SKU) can increase sales up to 20 times its baseline demand.  Even if Granarolo can successfully model and forecast the impact of these promotions, its competition will see them as periods of unpredicted demand decline that increases their demand volatility.

The increase of demand volatility in today’s markets explains why supply chain leaders tend to believe that their primary supply chain problem is forecast accuracy. They look for certainty of future demand from which they can make decisions and act. Yet, they cannot improve the forecast beyond its intrinsic variability. They can only deal with the variability through supply chain responses.

At that critical point of “hitting the ceiling”, understanding uncertainty becomes much more important and plausible than trying to increase forecast precision. That’s when supply chain leaders need to stop avoiding uncertainty and start embracing it!

A fundamental shift

In order to embrace uncertainty, the fundamental shift businesses need to make is to move from a deterministic model to a stochastic (probabilistic) model supported by appropriate tools, processes and people skills. A purely deterministic model is based on the premise of being able to know all the variables that can affect a business and therefore be able to predict the future with absolute certainty. In other words, a pipe dream!  The probabilistic model assumes there will always be a certain percentage of “known unknown” variables and attaches probabilities to them.

Here’s a simple example. Rather than deterministically stating that an average quantity of a given product will be shipped on a specific day, the probabilistic company models the probabilities that a customer will order a specific quantity on a certain day. So the order of 35 cases for delivery to a retailer’s Dallas distribution center received today had a 64 percent probability. Based on this modeled probability, holding an adequate inventory for the ordered item hedges the risk related to the demand variability and guarantees the planned service level.

The deterministic company waits for plans to blow up and then firefights. This causes all kinds of issues such as overreaction, cost overruns and service shortfalls, not to mention stress and sleepless nights. By contrast, the company that embraces uncertainty is prepared for it.

At this point you might very well be thinking, “Oh no, not more transformation.” But in reality most companies are already partly down the path towards a probabilistic supply chain model. Next week, we will explain that evolution.

Topics: Forecasting Demand and Analytics, Supply Chain Planning

Share this article or your comments below: