~ Contributed by Mike Gilliland ~

If we know the future demand for our products or services, we will only invest in the resources, capacity, materials, and staffing that will most profitably satisfy that demand.

But we don’t know the future, so we have to forecast…and that’s where the trouble begins!

• What is forecasting?
• How are analytics used?
• And what are the 3 areas I’d recommend you focus on?

What is forecasting?

Forecasting is the activity of trying to figure out what is going to happen in the future. It’s the first step in an organization’s planning process, and usually begins with a “statistical” forecast that has been generated with forecasting software. Of course, historical data may be unreliable or incomplete, or (for new products or services) there may be no historical data at all, so some manner of management judgment (often mischaracterized as “management intelligence”) is usually applied. Judgment can be applied through a simple override by a forecast analyst, or through an elaborate consensus or collaborative process that involves participants from throughout the organization (sales, marketing, finance, operations, and even executive management).

The end result of the initial statistical forecast and the ensuing judgmental overrides is a final “approved” forecast that feeds downstream planning systems. The forecast should represent an “unbiased best guess” of what is really going to happen (what we will ship or sell, or what services we will provide). Unfortunately, in the business context, the forecast more often represents the personal agenda of the individual (or department) providing or approving the forecast. This may be a target or stretch goal provided by upper management – to drive sales to reach that target. Or it may be a conservative or “sandbagged” number provided by sales – to make it easier to beat the forecast and achieve bonuses or other accolades.

Unfortunately, when you ask someone for a forecast, don’t count on getting an honest answer!

The analytics of forecasting

Forecasting is a subset of advanced analytics, which involves (per Forrester [The Forrester WaveTM: Predictive Analytics and Data Mining Solutions, Q1 2010, p.2]) “the identification of meaningful patterns and correlations among variables in complex structured and unstructured, historical, and potential future data sets for the purposes of predicting future events and assessing the attractiveness of various courses of action.”

The technical or statistical side of forecasting utilizes a wide range of econometric, time series, and specialized models, such as exponential smoothing, ARIMA, dynamic regression, unobserved components, neural networks, and intermittent demand models. An emerging area of interest is time series data mining, and an important application of this is in the creation and evaluation of new product forecasts.

While new and more sophisticated statistical methods are being developed all the time, the unfortunate reality is that fancier models do not necessarily translate to more accurate forecasts. A more sophisticated model can fit the history better than a simple model – in fact it is always possible to concoct a model that fits the history perfectly. But the role of forecasting is to generate good forecasts of the future, and “over-fitting” occurs when a model tracks what is essentially random or irrelevant behavior in historical patterns and projects it forward. For purpose of generating the most accurate forecast of the future, surprisingly simple models will often do better.

My top 3 recommendations

While forecast accuracy is at the top of everyone’s wish list, accuracy is ultimately limited by the nature of the behavior – its forecastability. For example, in forecasting Heads or Tails in the tossing of a fair coin, you will be correct 50% of the time over a large number of trials. You may wish to forecast correctly 60% of the time, but this is impossible given the nature of the behavior you are trying to forecast (the tossing of a fair coin). Since we can’t always achieve the level of accuracy we desire, it is best to focus our energies in these three areas:

1. Realize that you probably aren’t going to be able to forecast as accurately as you or your organization’s management want, and strive to understand what accuracy is reasonable to expect. It is reasonable to expect to forecast at least as well as a naïve model – something simple to compute like a moving average or random walking (using the last known actual as your forecast of the future). If all your statistical modeling and elaborate processes end up forecasting worse than a naïve model, something is terribly wrong with what you are doing!

2. Don’t waste resources pursuing unachievable levels of accuracy (like calling a coin toss 60% of the time). Instead, strive to beat the naïve model, and do this as efficiently as possible. Naïve models can be surprisingly difficult (and sometimes impossible) to beat, and the amount of improvement you can achieve over a naïve model may be small (often just a 10% reduction in error). The key is to not squander time and organizational resources pursuing unrealistic levels of accuracy. Good automated forecasting software, such as SAS® Forecast Server, can generate forecasts about as accurate as you can reasonably expect them to be, and do this very fast and efficiently with little need for human intervention. Let forecasters focus efforts on the “high value” forecasts that are most critical to the success of your organization, and let the automated software do the bulk of the work.

3. Don’t do the stupid stuff! Use the method of Forecast Value Added (FVA) analysis to identify (and eliminate) those practices that are just making the forecast worse. Download the white paper in the link above, or read more about FVA, worst practices, and alternative approaches to business forecasting in The Business Forecasting Deal.

Parting thoughts

Business forecasting is a cruel mistress. She breaks the heart of those who love it. She breaks the spirit of those who take it on as a career. And she breaks the bank at organizations that fail to do it competently. Be aware of the limits of business forecasting, and don’t let it break you.

For ongoing discussion of these topics (and the opportunity to contribute you own thoughts), I welcome you to follow my blog, The Business Forecasting Deal.