Dealing with Forecast Inaccuracy

No matter how much we want it, and no matter how hard we try, we can’t always achieve the forecast accuracy desired. Forecasting Heads or Tails in the toss of a fair coin gives the perfect illustration (being right 50% of the time is all we can achieve over a large number of trials). So what … does a company do when the desired (or required) level of forecast accuracy is not achievable? Curse, give up, forsake forecasting altogether, and curse again? Or, look for alternative approaches to solving the business problem?

In the past week I’ve come across two nice treatments of this topic. Bob Stahl, who I’ll be working with on the organizing committee for the 2012 International Symposium on Forecasting, wrote a piece on “Dealing with Forecast Inaccuracy” in the TF Wallace & Company online newsletter. Bob cites the work of Dr. David Orrel and the wonderful quote, “…the aim is less to predict the future than prepare for it.”

As organizations, and as forecasters, we have to deal with the reality that we aren't gods, and we don't have knowledge of future events. So rather than waste heroic efforts on impossible tasks, we are better served by acknowledging the limits of our abilities, and managing our businesses (and our lives) to accomodate the uncertainty.

In The BFD (the book), which Tom Wallace so graciously mentions in “Words From A Guy Named Mike” at the bottom of the newsletter, several “alternative approaches” to address the business forecasting problem are considered. When the standard statistical approach fails to deliver the accuracy desired, judgmental overrides and consensus / collaboration are the most common next steps. But even judgment has its limits, and a more effective approach may be demand smoothing — that is, making the demand forecastable. (Organizational policies and practices often add volatility to demand, making it more difficult to forecast than it should be.) Supply chain engineering approaches can improve flexibility and responsiveness, so that accurate forecasting is not as necessary. And a very effective approach is pruning — getting rid of extremely low volume and difficult to manage products that may be impossible to forecast and are probably cutting your profits.

In his webinar “Using Exception Management to Improve Demand Planning and Execution,” Alan Milliken of BASF shows how process control methods can be applied in business forecasting. Alan illustrates use of the Coefficient of Variation (CV) of a demand pattern to roughly determine its forecastability. He shows how to use CV to segment products into those that can be forecast with statistical methods, and those which cannot be forecast accurately, and to which different approaches should be applied. In addition to the alternatives mentioned above, another way to manage the “unforecastable” products might be to treat them as make-to-order or package-to-order (rather than make-to-stock). Alan's webinar is a preview of his presentation at the forthcoming IBF Supply Chain Planning & Forecasting Best Practices Conference in Orlando, October 24-26.