The Evolution of a Forecasting Department

Forecasts are judged on accuracy. At Warner-Lambert Company, in the Consumer Health Product Group (CHPG), as with any other business, if a forecast isn't accurate, it's worthless. The goal of the CHPG Sales Forecasting Department is to be the best in the industry. The building blocks include accuracy, customer service, inventory investment and continuous learning. And as you'll see, they don't come overnight. The consumer products business generated $732MM in 1992 sales. The products are divided into three main categories: oral care, upper respiratory, and women and skin care. The product line include the mouth rinses, Listerine and Cool Mint Listerine, Benadryl cold and allergy products, Lubriderm skin lotion and e.p.t home pregnancy test. Each of the businesses are highly promoted and subject to considerable seasonal swings, constituting a formidable challenge to the forecasting professional. WHY FORECASTING SYSTEM? The CHPG Sales Forecasting Department was established in the late 1980s to help improve forecast accuracy and customer service levels by (1) introducing objectivity in the forecasting process, (2) developing comprehensive forecast and market models and (3) facilitating timely and actionable communication between Sales, Marketing and the Manufacturing organization. However, the role of the forecaster, with no formal process or forecasting tools, was limited to attending forecast review meetings and performing ad-hoc analysis. Though a step in the right direction, these efforts fell short of fulfilling the department's objectives and management's expectations. A project was initiated to identify and implement a forecasting system, tailored to our line of products and logistics needs. Key here was a robust tool set (i.e., seasonality decomposition, promotional analysis and "what-if" capability), as well as the ability to integrate with the order processing and logistic systems. One area that was particularly challenging was determining what volume measurement to forecast (i.e., demand, shipments and billings). Billings and shipments were the traditional measurements, while demand was an abstract concept that many just paid lip service to. Simply put, a billing is generated when a customer order is recorded, regardless of the due date. Thus, if an order is processed in July with a September due date, it becomes a July billing. A shipment, on the other hand, is created when the order is moved out of the distribution center. In our business, demand is the measurement which is most customer sensitive. Demand is defined as when the customer requires the goods, which is not necessarily the same as when they are shipped, due to back-orders, as well as manufacturing and/or distribution delays. For example, if an order is placed with a requested ship date of July, but is not shipped until August, it is considered July's demand. Likewise, if an order is placed in July with a requested ship date of September, it is considered September's demand. Thus we chose demand. After one year of developmental efforts, the main components of the system were "completed" and we began training the marketing staff. As we proceeded with the training and roll-out plan, it became increasingly evident that the system was overwhelming to the users. This was due to: 1. The statistical and forecasting knowledge required to generate and interpret the models. The system provides eight "modified" trend analysis models, with a user override interface. This interface opens up a window in the models to account for shifts in the promotion calendar, as well as Sales and Marketing input. Although designed to minimize the need for heavy statistics, the user must possess a good understanding of moving averages, trends, seasonal decomposition, random noise, etc. However, marketing professionals are generalists and often lack statistical training. 2. Time investment by the marketing assistants. …