Automated forecasting software and managers who are involved in day-to-day activities each have their strengths and weaknesses. Extensive research, analysis and consulting leads Sanders, professor and Iacocca Chair at Lehigh University, to conclude that a combination of the methods is likely to ensure more accurate forecasts. The following interview took place in September, at the annual conference of the Council of Supply Chain Management Professionals. A video of the conversation is available at www.SupplyChainBrain.com.
Q: Why is forecasting so much more difficult than formerly?
Sanders: I've been doing forecasting for well over 20 years, I work with Nike and others, and I find it has become so much more challenging not just over the last 20 or 10 or even five years, but even just in the last couple of years.
First of all, everybody is global today. Our markets are global, our customers are global, our suppliers are global. So our span of control has dramatically increased. We also have very short product lifecycles. We have high expectations of very quick response times. A disruption anywhere across the globe affects us immediately, which did not happen even five years ago. So forecasting is much more challenging.
And you have to think also, we're not just forecasting demand. We're forecasting new markets, we're forecasting trajectories of those markets, the lifecycles of those markets. We're forecasting competition, we're forecasting whether markets will emerge and collapse. There's so much more to forecast and so many more factors that come into play. It's much more difficult than it's every been.
Q: Quite a number of automated forecasting packages purport to optimize the process. How effective are they?
Sanders: Most of the forecasting software packages are either stand-alone or tied to an ERP system of some sort. They all work off of quantitative forecasting models, and they're really the same kind of models we've seen and used for 20 to 30 years. We're technically still using the same kind of models albeit they are now part of a software package, and even though we live in a very different environment that has so much change.
One thing about quantitative models is that, they are only as good as the data they are based on. They are only as good as that data. We live in a time of change, so much of the historical data isn't applicable any more.
I work with companies, and we look at the guts of most of the forecasting engines, and so many are just based on the same quantitative models we've seen over the years, so I think for that reason it's problematic to be exclusively relying on the quantitative-model forecasting software.
Q: Do you have an example you can share?
Sanders: In fact, Nike, just a few years ago, in a highly publicized case - they had purchased an i2 software package, highly promoted, but what happened is that after they used the package, six months later they were in court. There was too much slow-moving inventory, not enough of the fast inventory that was needed. When they looked back, what they found is that managers were exclusively relying on the quantitative automated forecasting software and not enough on managers who actually knew about their products.
Q: So are you saying we should rely on managerial expertise instead?
Sanders: Here is the tricky thing. In academics, we often say that managers don't really have anything to say, we need to rely on quantitative methods. My experience is that that isn't really the case. Experienced managers who know their industry have a lot of insight. They come to conferences, they hear the buzz, they know what's happening. So, we really can't just rely on automated packages. We need to find a way to incorporate what managers know, what's the latest they've heard. A quantitative model or software package can't incorporate the very latest, what you as a manager have just found out in the last three minutes when you were on the phone with your vendor.
Q: So it's both technology and people that make for success?
Sanders: Absolutely. Keep in mind both approaches have their strengths and weaknesses. The trick is to combine them in some kind of way.
Quantitative methods are consistent. They are objective. They always give you the same results. They can process tremendous amounts of information and data. We as humans can't do that, but they're only as good as that data that they are based on.
A manager is biased. We're all biased; we've documented that. We can be tired on Monday morning or Friday afternoon, so our forecasting won't be the same. Our ability to consider a lot of factors isn't very strong. However, unlike an automated forecasting model, we are privy to a lot of insight into the industry that were in, so we need to find a way to harness the strength of both methods. That's really the way forward in terms of forecasting.
