The Business Forecasting Deal
By Michael Gilliland
John Wiley, 252 pages, $59.95
On the edge of a new year, one of the imponderables is how big demand will be for our products or services and how accurate our forecasts will prove to be.
If our predictions have been poor in the past, the hope usually is that we can improve them by investing more resources in the forecasting process. Perhaps we can generate more data, buy sophisticated new forecasting software, or hire more analysts.
But in The Business Forecasting Deal, Michael Gilliland, product marketing manager at SAS Institute Inc., tells us we can't control the accuracy of our forecasts. Some products have relatively stable sales and are easy to predict, but some are highly volatile and by their very nature will lead to seemingly botched forecasts. New products are a particular quandary, often without any comparable sales pattern upon which to base a forecast.
Instead, he suggests we focus on controlling the processes and systems we use and the resources we invest in the forecasting effort. "We should not squander resources in pursuit of unrealistic accuracy objectives that will never be achieved. Instead, we must prepare our organizations to deal with the uncertainty, not assume we can overcome it simply with forecasting," Mr. Gilliland advises.
Take what happens when you flip a coin repeatedly, which should be easier to calculate than predicting your sales for next year. We know the odds of it coming down heads or tails: 50 per cent. He asks you to forecast three processes: In the first, every day for 100 days the coin will be tossed 10 times; in the second, every day the coin will be tossed 100 times; and in the third, there would be 1,000 coin tosses a day.
Even when we know everything there is to know about the rules guiding the behaviour in the coin toss - unlike your customers' behaviour - the amount of inherent randomness limits how accurate we can be. In the first process, accuracy would be 77 per cent; in the second, 92.2 per cent; and in the third, even with 1,000 coin tosses for 100 days, the forecast accuracy is still only 97.3 per cent.
In real life, away from coin tosses, there's a lot more possible volatility. Most products and services have natural variations in consumption, such as the sale of more winter boots in winter, sunscreen in summer, and calls to service centres during the day rather than night.
But beyond that, systems of production and sales can create volatility. Salespeople trying to achieve their bonuses can artificially boost sales close to the end of each quarter. While consumer purchase of your product may be consistent from week to week, the companies that translate that demand into purchases may replenish their warehouses more irregularly.
Mr. Gilliland says the place to start reviewing your forecasting is by developing "a naive forecast" - a simple, easy-to-calculate prediction. Try a random walk, extending your last actual figure into the forecast, so that last month's sales become the forecast for next month. Or try a seasonal random walk, using the corresponding period from last year as your forecast. Another possibility is a moving average, taking an average of the past 12 months.
The naive forecast may be good enough for your purposes. If you are currently more sophisticated in your forecasting than that, the author asks you to evaluate how much value is added by your fancier forecasting efforts. Take each step in the process, and see whether it makes your predictions better. That will help you to identify waste and inefficiency on the forecasting process.
You may well find that certain features of your process - perhaps even the final massaging of the estimate by your CFO - actually makes the forecast less accurate than the naive forecast, so eliminate that step. Even if a step might improve value (perhaps a new computer modelling system some software developer is touting), is it worth the cost? And consider which elements of your business processes - such as quarterly bonuses to sales staff - may make sales demand more volatile, and harder to predict.
This book is a useful guide to forecasting for executives who want to gain a better understanding of their organization's efforts. The statistical elements are kept under control for non-experts, and Mr. Gilliland leads readers gently through the murky world of forecasts, illuminating the many pitfalls to avoid. It's ideal reading for the start of a new business year.
I'm Tempted To Stop Acting Randomly (Andrews McMeel Publishing, 127 pages, $14.99) is the latest compilation of Dilbert cartoons, with Scott Adams's delicious look at the office.
New marketing guru David Meerman Scott offers advice on an era where everything happens quicker in Real-Time Marketing and PR (John Wiley, 244 pages, $29.95).
Career adviser Donald Asher offers strategies to help you find a job in Cracking The Hidden Job Market (Ten Speed Press, 198 pages, $16.95).
Special to The Globe and Mail