“Forecasting is an ‘inexact science’,” writes Keith Peterson (@KPeteHalo), President and CEO of Halo, “that relies on the data available to you, the math you use, and how you implement the forecast.”[1] I doubt anyone would disagree with Peterson’s description of forecasting as an inexact science; but, being inexact doesn’t make forecasting unimportant. About a decade ago, reports Paul Taylor, Kimberly-Clark decided to transform its supply chain into a value chain encompassing consumers, retailers, and suppliers.[2] In order to accomplish that, Kimberly-Clark had to focus “on the information flows that link demand at the consumer level with supply capability, removing any complexities that may impede the ability to respond.” The resulting demand forecasting process helped the company reduce the number of plants it had to operate (from 19 to 15) and the number of warehouses it had to manage (from 80 to 25). The bottom line is that Kimberly-Clark was able to reduce its total supply chain expense and increase its gross margins. Those are impressive results; but, demand forecasting isn’t a silver bullet. In the area of inventory management, for example, Lora Cecere (@lcecere), founder and CEO of Supply Chain Insights, writes, “While improving forecasting sounds like the right answer, and companies need forecasting capabilities, what I see working today is not as simplistic as improving forecasting.”[3] In life, few things are simplistic, and getting demand forecasting right is no different.
Todd Taylor, a partner at OPS Rules Management Consultants, rhetorically asks, “Why do we put so much time and effort into improving forecasts? It’s probably because inaccurate forecasts lead to lost sales or inflated inventory, which in turn cost our companies lost revenue or unnecessary costs.”[4] Jeff Bodenstab and Stefan de Kok (@wahupaSCM) believe that too many companies don’t understand all there is to know about demand forecasts. They observe, “Maybe you are doing your demand forecasting completely wrong. OK, to be more precise, there are two equally important outputs of demand forecasting and you may be focusing nearly all your energy on only one, and maybe even the wrong one.”[5] They continue:
“And the impact is that you may not be getting the forecast accuracy you want. Or even more important, that you may not be getting the service levels and inventory efficiencies that you need. And if that’s true, you are not alone. The number of companies is growing that are saying that their forecast accuracy, service levels and inventory efficiency metrics have hit a ceiling that they just can’t get past. … Demand forecasts should predict two outcomes: the expected demand and how much uncertainty there is in that prediction.”
They go on to explain the circumstances in which focusing on expected demand is the right approach and the circumstances in which focusing on forecast uncertainty is more important. “If you have a lot of fast moving stock and unaggressive service level targets,” they write, “then focusing mostly on expected demand and forecast accuracy is probably a reasonable approach. … But if instead you are facing more long tail demand or aggressive service levels, then predicting uncertainty becomes much more important.” How, you might be asking, can knowing how much uncertainty there is in a forecast improve supply chain performance? Bodenstab and de Kok explain:
“Variability is perfectly normal. Surprisingly, while you can’t predict demand perfectly due to its inherent variability, you can predict demand variability. For example, you cannot predict the outcome of the role of two fair dice. There is just too much variability. But you can accurately predict the range of possible outcomes and the probability of each outcome. That is, you can predict its variability. And when you predict the variability, not only is this a more realistic task, it also leads to better business decisions. Consider a weather forecast. If you make an exact ‘rain’ or ‘no rain’ prediction, that information is helpful, but not nearly as helpful as providing a probability of rain. If you are planning an outside activity knowing there is a 5% chance of rain, you will likely make one plan. If there is a 40% chance of rain, you may make a different plan or have a backup plan for inside. That 40% figure provides you with more information to make better plans than an absolute rain/no rain prediction. It doesn’t take much to see that such a probabilistic forecast and improved prediction of uncertainty is useful information in supply chain planning. And in fact, accurately predicting uncertainty can add enormous value. That’s because you are focusing on improving not just to the average demand prediction but the entire range of possible demand predictions — including the extreme variability that has the biggest impact on service levels. Because when supply plans or safety stocks are based on wrong assumptions about demand uncertainty, targets go unmet and supply chains go into firefighting mode. And predicting uncertainty better leads directly to smarter safety stocks—able to capture a larger range of possible demand outcomes, catch more demand spikes, and buffer the rest of the supply chain. With a focus on the range of possible demand values, the number and the severity of supply chain disruptions decreases, leading to a more stable supply chain.”
Peterson adds, “Your forecasting success is fundamentally impacted by your understanding of that data, its strengths and limits.” Peterson offers recommendations about how companies can improve four data management concepts that should lead to better demand forecasting. The first thing you need to concentrate on, he believes, is preparing your data for analysis. “How you roll up your data for forecasting fundamentally impacts accuracy,” he writes. “In forecasting, the higher the level of aggregation the more accurate the forecast.” He concludes, “Understand the level of granularity required for your business purpose and the relative accuracy you can achieve at each level of aggregation.”
The next things Peterson suggests you concern yourself with are data currency, coverage, and accuracy. He believes “sales history is the best measure of future demand.” I’m not sure that is correct. A few years ago, Terra Technology published a study about forecasting in which it concluded, “The average error was 48 percent on a weekly basis, with ‘best’ performers cutting that number to 42 percent.”[6] If history was really a good indicator of future demand, one would expect the error rate to be much lower. I’m not saying that sales history isn’t important — it is — but it is only one variable that needs to be considered in a forecast. Peterson admits, “Before you decide on your models, you need to know whether your sales history is the same as demand for your product.” Todd Taylor explains that the timeliness of data also matters. “Optimize your inventory allocation process based upon an understanding of your demand certainty,” he writes. “If you have stable demand for some products, a push strategy can be employed. But where demand is uncertain, a pull strategy will need to dominate your policy.” Peterson concludes, “Create a measure of true demand by making an adjustment to account for stock-outs.”
The third thing Peterson suggests is learning to understand how order fulfillment can affect your forecasting. “Most businesses have situations where stock shortages occur but sales are fulfilled through alternate channels,” he writes, “such as an expedited order from a different location. The customer is happy but this can create chaos in the forecast because data collection can’t easily track this kind of exception.” He concludes, “Some slippage is going to occur in any large inventory system if inventory is not 100% on the shelf. To resolve this, simply record the place, time and item where the transaction actually occurred, along with availability.” As Cecere noted above, nothing is simplistic and manual processes are unlikely to be able to keep up with all of the variables required to make forecasts more accurate. Fortunately, cognitive computing systems are now maturing that can handle large number of variables and greatly increase forecast accuracy as a result.
The final thing Peterson suggests doing is managing “data spikes” that can skew forecasts. “Most businesses see occasional ‘spikes’ in their sales,” he writes. “Sometimes they are data errors, and sometimes they reflect real sales. Unfortunately, spikes tend to ‘pull’ the demand distribution in their direction, possibly skewing inventory planning.” A good cognitive computing system can help deal with spikes by understanding potential causal factors. For example, a devastating storm could create a spike for certain products but that spike would only have value if another similar storm were predicted sometime in the future.
Osgood Vogler, Director, Analytics, Celestica Supply Chain Managed Services, observes, “Implementing an insight-based demand management process structured around an understanding of key insights from human wisdom and analytical data is not a ‘set it and forget it’ decision.”[7] A truly useful demand forecasting system continually updates forecasts as conditions change. A supporting cognitive computing system can monitor necessary data inputs and provide actionable insights that can greatly reduce forecast error and even predict potential spikes for products (like those required following storms or catastrophes). At the end of the day, Peterson is still correct that forecasting is an inexact science, but a good cognitive computing system can place more emphasis on the science than the inexactness of forecasts.
Footnotes
[1] Keith Peterson, “Four Steps for Better Demand Forecasting,” Halo, December 2015.
[2] Paul Taylor, “Demand forecasting pays off for Kimberly-Clark,” Financial Times, 10 September 2011.
[3] Lora Cecere, “Does Better Forecasting Improve Inventory? Why I Don’t Think So Anymore.” Forbes, 29 November 2015.
[4] Todd Taylor, “Dealing with Inaccurate Forecasts,” Supply Chain Digital, 22 August 2012.
[5] Jeff Bodenstab and Stefan de Kok, “Here’s What’s Wrong with Demand Forecasting,” The Innovator’s Solution, 21 December 2015.
[6] “Using Demand Sensing to Boost Forecast Accuracy,” SupplyChainBrain, 21 February 2012.
[7] Osgood Vogler, “What If You Could Take The Guesswork Out Of Forecasting?” SupplyChainBrain, 20 June 2014.
