The demand forecasting process is the first step in supply chain management. Modern supply chains are getting more and more complex everyday. So it’s really important to forecast the future demand as accurately as possible which will make planning and execution of downstream processes easier and smoother .
Demand forecasting (forecasting for short) is the art and science of making predictions. “Science” because there is rigour involved in extrapolating historical data to make projections into the future. In many instances, data inputs that are forward-looking are as well incorporated using special algorithms in a process that is called demand sensing. The relevance of the “art of prediction” increases with decreasing relevance of historical data to the anticipated reality (what is to come). For example, under high uncertainty, it requires tact and expert judgment not to venture a prediction and wait until more data is available. The process of seeking consensus (especially when there is lack of clarity on what the data is telling us), for instance, requires skills that go beyond the quantitative rigour of the hard sciences and calls upon softer skills that belong in the social sciences. Art also requires one, as Paulo Saffo  puts it, to be open to the full range of possibilities instead of a “limited set of illusory certainties”.
The above description does give the sense that the process is not very straightforward. It might lead one to ask a fundamental question “why forecast in the first place?” It is not an irrational question, and the simple answer is that customers are unwilling to wait for as long as it takes for the supply chain to fulfil their request after a firm order is received. Therefore, it is necessary to start activities in anticipation of an order. The level of uncertainty and the length of the forecast horizon are rough measures of the extent to which activities have to happen in advance. The type of product that is forecasted (is it functional like a toilet paper or an innovative product) also plays a crucial role, and as do several other factors. Furthermore, besides the customer, there are resource allocation activities that need to happen more on a tactical (medium-term) and strategic level that requires plans far in advance of when decisions and actions of a more operational sort become due.
The reality for most supply chains is illustrated in Figure 1 (stock is built to forecast in anticipation of an order so that it can be immediately serviced).
Given the inevitability of forecasting in most situations, what are some of the key elements needed to excel? Many of the elements that are important for successful outcomes rely on making the right trade-offs.
The last point, and possibly the most important trade-off, is visualized in Figure 2. It shows that with decreasing model sophistication, the cost of inaccuracy climbs—there is an optimal region that balances the cost of forecasting and cost of errors that organizations should aim to achieve.
Modern tools for demand forecasting should cover the whole range of the methods mentioned here, from simple to complex, from "moving avarage" to the latest "machine learning" algorithms, in order to be able to reach the "optimal region" in the ratio of forecasting accuracy to forecasting effort for your company and your product portfolio.
The Pareto rule (e.g., 20% of products drive 80% of revenues) has long served organizations well in terms of determining priorities for planning. However, with the growing penetration of Internet and digital technologies, search costs (costs associated with the discovery of products) have decreased substantially leading to a corresponding increase in the share of sales of “niche” products—this phenomenon has been termed the “long tail” . The phenomenon emphasizes the already growing importance of differentiated treatment of products based on various planning relevant characteristics.
At a minimum, considering a value dimension like revenue or contribution margin (ABC[D]) and a volume dimension that captures the variability (of either actual demands or forecast variability) (XYZ) is essential to classifying products (commonly known as ABC/XYZ classification), which in turn is linked to planning strategies. Such an approach helps avoid the dual curses of planning efficiency: the “one size fits all” approach where salient differences between products are ignored and the other extreme where similarities across products are entirely discounted. In Figure 3, one approach to segmentation is shown where the planning segments drive decisions on organizational responsibilities, business targets, and level of automation (judgmental versus automated forecasting).
 Ideas adapted from: Kepczynski, R., Jandhyala, R., Sankaran, G. and Dimofte, A. (2018), Integrated business planning: How to integrate planning processes, organizational structures and capabilities, and leverage SAP IBP technology, Management for Professionals, Springer, Cham.
 Saffo, P., 2007. Six Rules For Effective Forecasting. [online] Harvard Business Review. Available at: <https://hbr.org/2007/07/six-rules-for-effective-forecasting> [Accessed 8 May 2020].
 Harvard Business Review. 1971. How To Choose The Right Forecasting Technique. [online] Available at: <https://hbr.org/1971/07/how-to-choose-the-right-forecasting-technique> [Accessed 8 May 2020].
 Brynjolfsson, E., Hu, Y. and Simester, D., 2011. Goodbye Pareto Principle, Hello Long Tail: The Effect of Search Costs on the Concentration of Product Sales. Management Science, 57(8), pp.1373-1386.