MRP/ERP

It’s Not the Fault of the Forecast

The word “forecasting” congers up all kinds of bad things for most of us; we’ve all heard that a forecast is “always wrong,” will maybe not always wrong, but never right. How can a company forecast in today’s world of instant gratification and customers deciding at the last moment on what they want, and then of course wanting what they want when they want it? And yet it is, if not impossible, very highly improbable to operate a manufacturing business without some sort of knowledge, information, indication, or forewarning of what is going to be required in the future.

Some years ago, I had the pleasure of being the Materials Manager for a major ball bearing manufacturer; bearing manufacturing at that time had a lead time of 13 weeks with a 26 week lead-time for the steel. With four categories of bearings (single row, double row, angular contact, and pillow blocks), three load bearing groups (light, medium duty, and heavy duty), and upwards of 15 to 20 individual sizes (from 1" to 36" in diameter), you can just imagine the chaos if we didn’t try to forecast future demand. With this kind of lead-time problem, we were required to forecast demand nine months out in the future. Forecasting was, needless-to-say, a very crucial part of operating the business, and required the joint involvement of sales and marketing, customer service, and planning and scheduling, with participation and strong support from purchasing and manufacturing. Were we always right in our forecasts? No! But we were far better off than trying to respond to demand only after being hit by it.

Too often we rely only on historical demand data to predict our future demand requirements, and for some industries this might be okay, but for most, the past is not a good harbinger of the future. The future is way too volatile to be predicted by the past.

The bearing manufacturer had been forecasting long enough to have learned that forecasting is a very dynamic function and must be constantly evolving. There are many factors that affect the demand of every business, and it is imperative that we recognize and factor all of them into our forecasting models.

At another company, we were purchasing gas directly from the ‘wellhead,’ and therefore were required to forecast our gas requirement one month at a time, fifteen days prior to the beginning of the month (June 15 for July’s fuel consumption). Being located in the northeast, seasonality played a very important part in the rate at which fuel was consumed each month.

Having some knowledge of forecasting for seasonality, I began by collecting, collating and analyzing monthly fuel consumption rates for the previous three years. The consumption rate curves for this data, starting with March as the first month of a year, was almost a perfect horseshoe; consumption begin to decline through July and August, and then begin to increase in September. For the next three years I rode the gentle horse of success; I would forecast accurately each month within the parameters. Exceeding our forecast resulted in the gas transporters to purchase more fuel and pay “spot market” prices; not consuming as much as was forecasted meant the transporter had to cancel gas purchases made to supply other customers. Both scenarios resulted in some form of additional charges to us, thus negating any savings we were to have realized from the direct purchase of the gas itself.

As I said, this lasted for about three years, then things started to change and my forecasting knowledge came up short-handed. The maintenance department had been going around and replacing all the broken windows throughout the entire facility and caulking all the windows and doors, doing a great job of keeping the cold winter air out of the building. This in itself could have been compensated for by shortening the number of years in the ‘averaging period,’ in essence increasing the sensitivity and volatility of the forecasting. The straw that broke this camel’s back was Mother Nature. The old girl started mixing up things; winter weather became less predictable, with one month warm and maybe the next cold. Here is where my limited knowledge and homegrown computer software program was just not up to the problem.

The forecast cannot be blamed for giving inaccurate information; it was my limited knowledge which resulted in not providing the all the influencing factors to a software program capable of using the data and generating good, useful, and usable information.

Today, forecasting not only employs the old, familiar techniques of moving average, weighted moving average, exponential smoothing, and regression analysis, it has recently brought to bare the power of simulation. The National Weather Service has been using this technique for many years; each time a new weather condition develops, the Service uses a simulation model and plays the “what if” game. With each passing hour, new data is collected and fed into the simulation model and the game of “what ifs” is played once again. We may still complain about the weather and the accuracy of the forecast, but we must all agree that weather forecasting has gotten more reliable over the years and we do tend to depend on it, whether we like to admit it or not.

Along with the power now available in the modern desk top computers and the new software written for them, this technique is available for relatively few dollars. And the impact this has on more accurately allowing you to forecast the future for your business impacts the bottom line directly.