13.1 Forecasting Organization
Much of this book has treated forecasting as a predominantly statistical exercise—making use of information within the organization to reduce demand uncertainty. In practice, forecasting is as much an organizational exercise as a statistical one: Information is diffused throughout the organization, and the forecast has many stakeholders all of whom require the forecast as an input to their planning processes. Understanding demand forecasting requires an understanding of not only the statistical methods used to produce a forecast but also the process of how an organization creates a forecast and uses it for decision making.
Managing this process well is the realm of Sales and Operations Planning (S&OP). Given that S&OP has been around for a while, best practices for such processes are now established (e.g., Lapide 2014). There is an input and an output side to this process. On the input side, a good S&OP process supports sharing relevant information about the demand forecast. The emphasis here is on marketing and sales to share upcoming product launches, acquisition of new customers, planned promotions, and similar information with those responsible for forecasting to help them in the process. At the same time, other functions such as operations need to share relevant input data into the planning process as well, such as the inventory position, capacity available, and so forth. On the output side are a set of coordinated plans that all use the same information as input. Marketing develops a plan for promotions and demand management. Operations develop production and procurement plans. Finance develops cash flow plans and uses numbers, in coordination with the other departments, to communicate with investors. Human resources develop a personnel plan based on the same forecast data. If this process goes well, the organization will be coordinated, and decisions and plans will be based on all available information. If this process does not go well, information will be hoarded, forecasts will be influenced by individual members, and different functions of the organization lack coordination. The result will be highly inaccurate forecasts and, as a consequence, promotions that may not be backed by capacity and forecasts shared with investors that bear little resemblance to actual plans of the organization, thus damaging the firm’s credibility.
An S&OP process is a monthly process within an organization to enhance information sharing and coordinate plans across the organization. It usually involves a cross-functional team from marketing, operations, finance, and sometimes human resources. There are typically five different steps in an S&OP process. Beginning with data gathering, representatives from different functions share relevant information with each other and develop a common set of business assumptions that go into the forecast. In the actual demand planning stage, the team finalizes promotion and pricing decisions and agrees on a consensus forecast. This forecast is then used to develop sales targets for the sales function. Afterwards in the supply planning stage, inventory, production, capacity, and procurement decisions are made according to the forecast. Further, if shortages are expected, rationing and prioritization policies are developed; if major risk events are considered, contingencies are developed and shared with all members of the team. Finally, in the pre-executive meeting, senior management can adjust any outcomes of the S&OP process. The finance function, in particular, sometimes has a veto power at this step, to enable better cash flow planning and investor communications. Finally, all relevant plans are discussed and finalized with top management during the executive meeting.
13.2 Organizational Barriers
There are two essential barriers to overcome in order to make this process work: incentives and organizational boundaries. The former barrier stems from the fact that incentives across different functions are not aligned, and often no one is held accountable for the quality of forecasts. Managers in marketing or sales may have an incentive to lowball the forecast, since they understand that their sales targets are often set depending on the forecast, and thus lowering the forecast is an easy way to make their targets more obtainable. Or they may have an incentive to inflate the forecast, since they understand that this will push operations to make more products available, thus decreasing the chances of a stockout happening and thereby increasing sales-related bonuses. Managers in operations, on the contrary, are often compensated based on costs; one way to keep less inventory is to lowball the forecast and thereby lower production volumes. Finance will chip in here as well, since they will use the forecast to manage investor expectations. These incentives will inevitably lead to a confusion of forecasting and decision making; the actual forecast thereby becomes not one number to coordinate all activities but a playball of organizational politics.
The latter barrier is a result of different functional backgrounds and social identities. On a very basic level, people in marketing may have studied different topics than people in operations; they may have had a different entry route into the organization as well. As a result, their way of perceiving and communicating about organizational realities may be very different, leading to challenges in managing a cross-functional team involving both groups. The units they forecast may also be different—whereas finance forecasts revenue in US dollars, marketing may predict market shares and operations is interested in units of production. While these units can usually be easily converted, they represent a natural barrier to overcome, and the conversion of these units should be standardized upfront. Last but not least, having different functions always implies different social identities that are difficult to overcome. Representatives from marketing will feel a natural allegiance to their function, as will representatives from operations. While such identification usually creates more trust within the group, it creates distrust between groups. Successfully managing a cross-functional S&OP team thus requires not only establishing standardized norms of communication, it also requires breaking down functional barriers to build trust between representatives of different functions.
One approach to solve these barriers is to create a forecasting group that is organizationally separate from all other participating functions; this promotes accountability for forecast accuracy. It also creates professionalism and allows people associated with this group to be compensated on forecasting performance. A survey on forecasting practice found that 38 percent of responding organizations have introduced a separate functional area for forecasting, and 62 percent of those separate functional areas owned the forecasting process (McCarthy et al. 2006). An essential advantage is that forecasters in this area can be compensated based on the accuracy of forecasts without creating asymmetry in their incentive systems; incentivizing forecasting in all other areas requires very careful calibration to offset the existing incentives in these areas to over- or underforecast (Scheele, Slikker, and Thonemann 2014). If creating a separate function is organizationally impossible, then the people participating in S&OP planning should de-emphasize their functional association. For example, it may be possible to completely take the participating employees out of their functional incentive systems and reward them according to firm performance or according to forecasting performance instead.
Another essential aspect is to demystify the forecasts; all those involved in the forecasting process should be familiar with the data, software, and algorithms used in creating forecasts. Assumptions made should be documented and transparent to everyone involved, and individuals should be held accountable for their judgment. As we emphasize in Chapter 11, it is ex post always possible to tell whether adjustments made to a statistical forecast were actually helpful or harmful to forecasting performance if one takes a longer term perspective. As such, an interesting avenue to resolve incentive issues in S&OP may be to use past forecast accuracy coming from different functions to determine the future weight that is given to these different functional forecasts when calculating a consensus forecast. Functions that bias their forecast will thus quickly lose their ability to influence the consensus forecast, creating an incentive for them to avoid unnecessarily biasing the forecast.
A good case study of transforming an S&OP process is given by Oliva and Watson (2009). The authors study an electronics manufacturer that started off with a dysfunctional forecasting process; the company had three different functions (sales, operations, and finance) creating three different forecasts; the only information sharing happened in nonstandardized spreadsheet and hallway conversations. The company proceeded to generate a process that started by (1) creating a separate group that was responsible for the statistical side of forecasting, (2) creating a common assumptions package where each function would contribute their key information about the development of the business, (3) allowing the different functions to generate separate forecasts based on the same information but then (4) integrating these forecasts using a weighted average, where the weight attributed to each function depended on past accuracy, and (5) limiting revisions to this initial consensus forecast to only those instances where real data could be brought up to support any modifications to the forecast. The result was a stark increase in forecasting performance. Whereas company forecasts had an accuracy (1-MAPE) of only 50 percent before this redesign, this accuracy jumped up to almost 90 percent after restructuring the S&OP process.
13.3 Key Takeaways
• Forecasting is as much a social and an organizational activity as a statistical one.
• Employees’ functional background may incentivize them to influence the forecast in ways that are not optimal for the business as a whole. Consider changing their incentives to align with getting an unbiased forecast.
• It may be very beneficial to organizationally separate forecasting out from traditional functional areas.
• Similarly, employees’ background may influence the way they think and communicate about forecasts. Keep this in mind and work for open communication.