The ultimate role of demand forecasting is its ability to help in decision support. The demand forecast information is the first step in enabling the making of sound business Figure 3. Tracking demand based on existing customer behavior
Existing
Existing orders – a mix of on-line and
Existing orders – a mix of on-line and
traditional channels
decisions that have a ripple effect on all strategic activities down the supply chain.
Decision support efficiency can be achieved by storing all relevant target customer behavior information in a data warehouse and by using data-mining software to extract analytics to help decision support. A data warehouse helps with the tracking of historical customer behavior, and the analytics help with understanding its relevance to current customer demand.
As shown in Figure 4, for a successful implementation of demand forecasting, the system’s activity must be supported by business processes, such as Consensus
Forecasting, Vendor Managed Inventory (VMI) and Collaborative Planning, Forecasting and Replenishment (CPFR).
This will reflect a true implementation of an environment where the systems support the people and not vice versa. Those people who are the decision makers should ensure that the factors affecting demand forecasting are integrated into every subsequent process from design through manufacturing, supply chain, Customer Relationship Management (CRM) and even on to customer feedback. None of these support tools is new, and we don’t propose to review them in this chapter; however, it is important to consider how they interact with new processes that are becoming available, as we move into the age of the Internet and advanced computer tools. Some changes and challenges are very obvious, and some are not so obvious. For now, let’s just consider consensus forecast-ing, which, in its simplest terms, just means a forecast that has been agreed to by more than one party. Originally, this meant two or more people getting into the same room and talking until they agreed on a number. As various tools were developed, it became possible for the people to work from remote locations, at differing times, and even on different tools that then were able to transfer information back and forth. As we move into the future, the big change in consensus forecasting will be in how we take advantage of new communications tools on the Internet in order to enhance the forecasting process.
As the overall processes become more and more complicated, it becomes more difficult for participants to stay abreast of the bigger picture. The bottom line of the consensus Figure 4. Integrating demand data into a forecasting process
Predicted
Data Mining Synopsis for Decision
Support
Supporting Processes
Consensus Forecasting Vendor Managed Inventory (VMI) program Collaborative Planning, Forecasting and Replenishment(CPFR) Predicted
Data Mining Synopsis for Decision
Support
Supporting Processes
Consensus Forecasting Vendor Managed Inventory (VMI) program Collaborative Planning, Forecasting and Replenishment(CPFR)
forecasting process is to gather various forecasts for the same things that are generated using different processes and then combine them to come up with a combined forecast to which all participants can agree. Most of the participants have real jobs to deal with and cannot spend large amounts of time forecasting. For example, a salesman’s main (or real) job is to sell, and we need salesmen to be out doing that. While forecasting is important and does add great value to salesmen, there is a definite limit to how much time they can allocate to the process. However that time is not only necessary, but it is also critical to the accuracy of the overall forecasting process. The salesman is the one that is out talking to the customers and has the best feel for what is going to happen in the future with those customers. It is easy to statistically predict future sales based on past sales, but that is a lot like driving using only the rear-view mirror. So long as the road is straight, there is no problem, but at the first curve in the road, you will be in the ditch.
When you try to forecast using only past sales, the ditch you run into might be only a slight inaccuracy, or you might drive right off a cliff, if there is a big enough change in the customer’s behavior. The point is that everyone’s opinion on future sales is
important, and combining them together appropriately gives you a better chance at improving your accuracy.
So, what will the forecast process of the future look like? In a lot of ways, it will look much like the process of today but with many additional levels and tools involved in the
process. If the virtual corporation is the model of the company of the future, then the virtual forecast is the model for demand planning. In the virtual corporation, you have many small groups or companies banding or networked together in order to accomplish some set goal. Each of the players brings something unique to the mix, with the actual combinations changing with time as the goals and needs change. Each of these small groups represents a node within the larger grouping. In virtual forecasting, you also have a nodal setup, with each node representing some logical subset within the larger context.
For example, you could have a different node for each global region. Within these nodes could be subnodes for each country, and within these nodes could be subnodes for some specific grouping, such as customer groupings. As shown in Figure 5, each subnode would work independently of other subnodes to the point of potentially using different tools and processes to generate the required demand requirements.
This process is not all that different than the situation today. Where the difference comes in is in tying each of these different nodes together, using net technology to allow an organization to roll up the demand to get a holistic view of total requirements, then adding in feedback mechanisms in the actual sales and distribution levels to real-time monitor and modify the demand requirements forecast, based on what is actually happening rather than the best guess that the forecast represents. Those organizations that can spot and react most quickly to unanticipated changes in demand are the organizations that will have a competitive advantage in the modern global marketplace.
When thinking about nodal forecasting processes, keep in mind that the key to subnode forecasting is the communications linkage between each of these subnodes. It’s like the old comparison among data, knowledge, and intelligence; the subnodes provide data, but without some way of making a synthesized whole or being able to easily find one specific piece of data, it is of no use. The problem, of course, quickly becomes a technological issue, when decisions need to be made on whether this gets controlled by one centralized processor or many little processors that communicate to each other.
Obviously, in the early days of computers, the first inclination was to centralize everything, and this attitude still heavily influences much thought and practice today.
With personal computers becoming more and more ubiquitous and ever more powerful, this may not be the ideal method anymore. Nicholas Negroponte (1996) talks about the concept of mass communication of many small processors in his book, Being Digital, where he makes the point of how much simpler it is to coordinate many small processes than most people would be willing to believe. By using a swarm of small computers to do the initial filtering and loading of data into the process, we gain far more information and far more useful information than we could with a single dedicated computer.
Another major factor in the forecasting process will be, again, at the subnode level, where new Web-based intelligence agents will be gathering information about target customers and factoring them into the overall forecast as an integral part of the consensus
forecasting process. No longer will the consensus process be just people interpreting data and reaching a joint conclusion, but instead, you will have computers taking a larger role in the actual decision process, not just in the data gathering and sorting processes.
Does this mean that computers will replace people in this process? Anything is possible, but until computers can be developed that have something equivalent to intuition, it is highly unlikely. While there is a large science component in forecasting, there is also a large art component, and it is this art component that still requires people in the process.
What this means is that what the forecaster of the future does will change, as we take advantage of what computers do really well—crunch data. By that, we mean the computers and the new tools that are becoming available are really good at sorting through large databases, at doing many calculations, and at reporting information according to the directions they are given. What the forecaster will be concentrating on is supplying those directions to the system and interpreting the information, once the computer has presented it in a defined and understandable format.
Figure 5. Map of virtual forecasting process