Dan Gilmore, Editor-in-Chief of Supply Chain Digest, has stated that his favorite vision of the future supply chain is one “where, when a sweater is sold in Peoria, somewhere in New Zealand a sheep is shorn.” That is the perfect demand driven supply chain. As Gilmore admits, however, “It doesn’t work like that yet today.” [“What to Tell Students and Bureaucrats about Logistics?” 11 February 2011] In the absence of a perfect demand and respond system, forecasts still play an important role in manufacturing and supply chain operations. Sandeep Deolekar, Lead Consultant for Enterprise Solutions at Infosys Technologies, asserts that forecasts help address the “following key challenges most of the Supply chain professionals face across multiple dimensions” [“Is ‘Higher Forecast Accuracy’ the silver bullet?” Supply Chain Management, 5 October 2010] Those challenges include:
1. Globalization — “The drive towards Globalization has resulted in the focus to not only look at the developing markets for cheap supply, but also to tap these developing markets to drive future growth.”
2. Extended Supply Chains — “Increasing lead times and lead time variability with most of the manufacturing bases of suppliers outsourced or offshore.”
3. Informed Customers — “Increasingly demanding customers with information at finger tips (thanks to [the] internet) and lower brand loyalty.”
4. Increased Competition — “Intense competitive activity driving lower prices and reduced scope for differentiation.”
5. Rapid Innovation — “Increased pace of product innovation – rapid new product introductions combined with rapidly reducing product life cycles.”
Deolekar insists that these challenges highlight “the need for Forecasting and Demand Planning … to ensure a steady flow of products to the right place at the right time in the right form.” The purpose of forecasting, he states, “is to identify some pattern or relationship amongst the relevant variables and then extrapolate from these the future.” The problem, of course, is that the further out into the future you try to forecast the less value extrapolation becomes — a point Deolekar makes below. Regardless of its drawbacks, pattern recognition is very important and technology can help discover patterns — even subtle ones. The better the accuracy of a forecast the more helpful it is. To achieve the highest levels of forecast accuracy, Deolekar asserts that you should consider the following assumptions (which he calls principles):
“1. The future is never the same as the past and hence a straight extrapolation will not always be helpful.
“2. The available historical data contains the underlying pattern inter-mixed with the noise (random element).
“These principles imply that the goodness of fit of the models to historical data has little correlation to the potential of the accuracy of the forecast generated by these models. Also complex statistical forecasting models run the risk of over-fitting to the historical data and mistake patterns for noise.”
So where are we left? Having tried to convince us of the value of forecasting, Deolekar then proceeds to raise questions about it. He insists that he is not undermining the case for forecasting; rather, he is trying to point out that a degree of uncertainty is inevitable. Drawing from a book entitled Dance with Chance: Making luck work for you, written by Spyros Makridakis, Robin Hogarth and Anil Gaba, Deolekar writes:
“[Referring] to the approach [in …] ‘Dance with Chance,’ … the authors propose the following 3-A framework to deal with Uncertainty:
1. Accept that you are operating in an uncertain world and thus identify the range of possibilities.
2. Assess the level of uncertainty using the available data and any additional inputs.
3. Augment the range of uncertainty estimated in the earlier step.
Applying this in the context of Demand Planning, we must not only generate statistical forecasts, but also assess the uncertainty involved in the demand patterns and further augment this uncertainty to create a range of forecast.
The benefit of using “a range of possibilities” is that flexibility is inherently part of the process. In a subsequent post, Deolekar talks about the famous “Pareto’s Principle or the 80/20 rule or the ‘Law of Vital Few’.” [“Forecaster ABCs – The ‘Vital Few’ for Forecasting,” Supply Chain Management, 23 February 2011] His point is that the most important forecasts to get right are those that concern the customers most vital to your business as well as forecasts about the products that generate the greatest profit for your business. There are a lot of ways to slice and dice forecasts and technology is helping make all sorts of forecasts easier and more affordable. As an example, Bob Ferrari posted a blog about a company name Lokad that provides mid-market and larger firms with a “cost affordable alternative in generating more timely and accurate forecasting.” [“A Twist in Cloud Computing- Forecasting Mathematicians On-Demand,” Supply Chain Matters, 10 December 2010].
Deolekar and Ferrari talk about the importance “accurate forecasting,” but supply chain analyst Lora Cecere laments that, even though we may have made some advances to improve forecasting, we have “not [made] a great leap forward.” [“Trading Places,” Supply Chain Shaman, 28 February 2011] She explains:
“[For one group of companies,] the average monthly Mean Absolute Percentage Error (MAPE) for a one month lag was 31% + 12%. Data eight years ago for the same companies was an average of 36% + 10% MAPE. The result? This group of consumer products leaders has gotten slightly; but not significantly better in demand forecasting.”
Rather than being appalled at these error rates, Cecere states that she “expected the results to be FAR worse.” Why? Because the inputs that go into forecasting have undergone a “storm of market changes.” Among these changes, Cecere includes: “product lifecycles, product proliferation, higher levels of promotions, changes in competitive behavior, and global expansion.” Commenting on Cecere’s post, Trevor Miles asks, “Is Forecasting Fatally Flawed?” [The 21st Century Supply Chain, 24 March 2011] He writes:
“So much of actual demand is unplanned. Which is fine as long as it is near to what was expected in terms of items purchased, period in which purchased, and the customer/region in which the purchase took place. But this does not appear to be the situation in many cases. So is forecasting fatally flawed? [Lora Cecere’s post] made me sit up and listen. Especially when she went on to quote from her research while at AMR Research that
Based on AMR Research correlations, a six percent forecast improvement could improve the perfect order by 10 percent and deliver a 10-15 percent reduction in inventory.
“In other words, there is a lot of benefit to getting the forecast right. But a range of highly respected CPG companies cannot do better than 31 percent MAPE, with a range of 19 percent to 43 percent? That caught my attention. Mostly because I am more familiar with the High-Tech/Electronics industry which has much shorter product life cycles than CPG and therefore more volatile or variable demand patterns.”
I think the real question Miles is asking is: Is flawed forecasting worse than no forecasting at all? His answer is, “No.” He explains:
“I hear from [High-Tech/Electronics] companies that they seldom get their forecast accuracy, as measured by MAPE, above 50 percent, which is consistent with my observations about the characteristic differences with CPG. Higher demand variability/volatility would imply a lower forecast accuracy. Before anyone jumps down my throat, especially Lora, let me state unequivocally that everyone MUST forecast and that all companies should be demand driven.”
I guess that means that forecasting is flawed, but not fatally. As Miles states, he isn’t trying to stop companies from forecasting; rather, knowing that forecasts are inherently inaccurate, he believes companies need to concentrate on being more responsive. He concludes:
“Where is the discussion about how best to satisfy the missing 31 percent demand in the case of CPG and 50 percent in the case of High-Tech/Electronics? Where is the discussion about the profitable response to the demand that is not anticipated? I feel as we are only having half the conversation. The half about forecasting. But if the best we can do is improve forecast accuracy from 64 percent to 69 percent over eight years in an industry segment with relative stable demand, I think we should be talking about supply chain agility and responsiveness. What amazes me is that since the early 1990’s we have been applying optimization engines, typically Linear Programming (LP), to the supply side. Ignoring for the moment the inherent issue of using linear models to represent highly non-linear systems, if you are basing your optimizations on inputs that are best 69 percent correct, are you not focusing on the wrong problem? Should you not be focusing on systems that enable you to detect true demand early and determine the best way to satisfy the unanticipated demand using the competing requirements of profitability and customer service? Of course you will need a supply chain that can execute in an agile and responsive manner consistent with your decision. Here is the rub: All our resources are limited. Time. Cash. People. So in this zero-sum game, where are you going to apply your energies? Spending eight years to improve the forecast by five percent, or working on the manner in which you satisfy the unanticipated demand in the most timely and profitable manner?”
I think he makes a great point. As noted earlier, forecasts are generally based on pattern recognition and extrapolating those patterns into the future. We all know, however, that consumer tastes change, new technologies emerge, and financial circumstances alter. These events create shifts for which extrapolations can’t account.
Up to this point, I have been discussing forecasts that affect operational planning. But forecasts can also affect strategic planning. Extrapolation is also often used to generate strategic plans. Unfortunately, so-called “black swans” make pattern extrapolation fraught with uncertainty. Ann Grackin, from ChainLink Research, insists that just because some events are rare that doesn’t mean we should not forecast them. That may sound oxymoronic, but Grackin writes, “Creating a resilient enterprise is critical to customer protection and employee welfare, as well as securing the financial viability of the company. Yet many firms think that since rare events are unpredictable, then there is no sense in doing much about them other than ‘risk transfer’ (purchasing a risk product, if one is available, such as product liability, property and casualty and so on). … Modelers and forecasters look through a faulty lens; they discount these events because they are rare.” [“Black Swan? Hardly! Revolutions and Tsunamis Come and Go!” 5 April 2011]
Drawing from the work of Nassim Nicholas Taleb, author of Black Swan, Grackin goes on to argue that forecasts can take into account rare events using history and a little common sense as a guide. She concludes:
“Crying out ‘Black Swans’ may be chic, but many of those who do may not understand Mr. Taleb. He challenges the notion that these are random and therefore unpredictable events, so we can’t plan and predict better to avoid being impacted. His perspective is that they are not as random as you think and that a lot of common sense—and evidence in plain sight—would lead people to more enlightened behaviors and preventions.”
Although analysts recognize that forecasting has its limitations, they also recognize that it has important uses. Lora Cecere makes us wary that forecasting will improve dramatically; but with more powerful computers and better forecasting algorithms being created we can only hope that baby steps turn into great strides.
