Launch II: How much will sell?

In Chapter 4, I described one way to forecast sales for a business case. Estimated take rates from a discrete choice study are applied to an estimate of the size of the target market to estimate expected total sales. This may be sufficient for a business case, but not at launch time for three reasons. First, once the product is launched, its performance must be tracked, as I discuss in Chapter 7, but part of tracking is assessing performance relative to objectives, otherwise you cannot tell if the product is doing well or not. Objectives are typically in terms of market share, return on investment (ROI), or sales growth. The sales force then becomes responsible for meeting these objectives. A forecast of units sold is the basis for these objectives.

The second reason for a forecast is that the manufacturing division of the business must produce the product so it needs to know how much to produce and how to plan for that production. Raw material must be ordered, factories must be retooled or new ones built, and personnel must be hired and trained or robots must be designed, built, and installed. Not only does this all require a lead time, but also the scale for this new capacity must be known. How much factory space is needed? How many new robots are required?

The third reason for a forecast is that final pricing has to be set but price points depend in part on expected sales; that is, demand. If sales are projected to be weak during an initial period after launch, then a low price point reflecting a penetration pricing strategy is required. On the other hand, if sales are projected to be extremely strong, especially because of early adopters, then a high price point is needed. More formally, in the former situation, demand is elastic so a low price point is needed while demand is inelastic in the latter situation thus dictating a high price point. All of this requires a sales forecast prior to launch. See Paczkowski [2018] on elasticities and pricing analytics and Nagle and Holden [2002] on pricing strategies.

In this chapter, I will discuss some of the mechanics and issues associated with developing a forecast. This will be done in eight sections. The first section

This Venn diagram illustrates the relationship between a forecast and a prediction

FIGURE 6.1 This Venn diagram illustrates the relationship between a forecast and a prediction. Predicting is a general concept with forecasting as a special case. This diagram shows that all forecasts are predictions but not all predictions are forecasts.

distinguishes between forecasting and predicting. They are not the same although they are nonetheless closely related. Forecast development responsibility is briefly mentioned in the second section. Since a demand forecast is based on a time series, it is helpful to define a time series. I do this in the third section. Data issues, which determine the possible type of forecast, are discussed in the fourth section. The fourth section provides some detail on forecast methods applicable for the amount and type of data available. Forecast methods and forecast error analysis are discussed in Sections 5 and 6, respectively. Sections 7 and 8 are for software and a summary. A technical appendix of the forecasting methods rounds out this chapter.

Predicting vs. forecasting

Let me first clear up confusion between the words “prediction” and “forecasting.” The two terms have similar meaning in that they refer to producing a number for an unknown case or situation. Basically, they both fill in a “hole”, something that is missing, in either our data or understanding. In this sense, they are the same. They differ, however, regarding the nature of the hole. Forecasting is concerned with time series data so it is concerned with saying something about what will happen in the future. The data “hole” is a future time period, a hole that should be obvious. Predicting, however, is concerned with an unknown case which could be in the future or it could be now under different situations. So you forecast this year’s sales but you predict what sales would be under different price points. The relationship is illustrated in Figure 6.1.

Forecasting responsibility

Sometimes, sales forecasts are developed by different organizations. A forecasting department and the sales department, for example, might independently develop

Forecasting requires a process just as the overall NPD requires a process

FIGURE 6.2 Forecasting requires a process just as the overall NPD requires a process. This flowchart illustrates how a forecast, developed by a forecasting organization which is responsible for developing and maintaining a forecast, is reconciled with views held by the sales organization. The resulting reconciled forecast is shared with other organizations that depend on it for their operations.

forecasts but the business can only use one. These diverse forecasts must be reconciled to produce one official forecast that guides manufacturing, sales, procurement, capacity planning, and pricing. A stylized sales forecasting reconciliation process is outlined in Figure 6.2.

Time series and forecasting background

Regardless of how a forecast is developed, a time series is needed. Let Vj, Y2,..., YT be a time series from period t = 1 to period t = T where T is the most recent, last period before the forecast begins. A forecast is made in period t = T for period t = T + 1 based on, or conditioned on, all the past history in the time series. For one method, described below, the history is of no consequence; just the last data point counts. This is obviously a naive method. Technically, the conditioning should be explicitly stated but I will drop reference to it for simplicity in what follows.

A forecast for a future period is referred to as a “step ahead.” A forecast for one period into the future is a “1-step ahead” forecast; a two-period forecast is a “2-step ahead” forecast; a forecast h = 1,2,... periods into the future is an “h- step ahead” forecast. A forecast made at t = T for the next period outside the history of actual observations is denoted as YT(1). This is a 1-step ahead forecast for period t = T + 1 made in period T (conditioned on the past). The h-step ahead forecast, YT(li), is the h-period ahead forecast made in period T. The set of values {Yr(l), Yr(2), YT(3),..., YT(k)} is a forecast profile. When a new value, YT+h, becomes available in period T + h, this value is referred to as an actual. There is a forecasted value for period T + h, YT(h), and an actual value, YT+h.

Data issues

Forecasting in general is difficult regardless of the responsible organization or the extent of a reconciliation process. The newness of a product compounds the difficulty because of the lack or sparsity of data. By definition, the product is not yet available to generate sales data. Nonetheless, there may still be data that can be used as I will discuss shortly. There are several methods available to forecast for products with some data history. These include econometric models, Autoregressive Integrated Moving Average (ARIMA) models, simple trending, and smoothing techniques such as exponential smoothing. Which one is used depends on the level of sophistication of the business analysts, the amount of data available, and the influence of seasonality and external economic patterns (e.g., the business cycle). See Wei [2006] and Hyn- dman and Athanasopoulos [2018J for discussions of general forecasting methods. Even for these methods, sample size is an issue since the historical data may still be insufficient. Mik [2019] and Hyndman and Kostenko [2007J note that a sample size of 16-17 months is necessary for ARIMA and exponential smoothing models to produce adequate and acceptable forecasts. Mik [2019] also notes that when seasonality is present, even more data are required. These sample sizes may not be, and often are not, possible for new products.

The type and amount of historical data depends on the nature of the new product. Not all new products have the same form of “newness.” Kahn [2006], reported in Mik [2019], listed seven types of new products based on their combination of embedded technology and market focus. Table 6.3 summarizes the combinations

Definitions of new products based on Kahn [2006]

FIGURE 6.3 Definitions of new products based on Kahn [2006].

and the seven types of “newness.” To simplify my discussion, however, I will only distinguish between two types: new-to-the-world (NTW) and not new-to-the- world (NNTW). The former are revolutionary items that have never been seen before in any form and people have no inkling they are about to be introduced to the market. In fact, these products create new markets. The Apple iPad, iPod, and iPhone are excellent examples of NTW products that had never been seen before and resulted in the creation of new markets which were eventually populated with a host of similar products. NNTW products, like the ones that followed the three Apple products, are basically variations of a theme of something that already exists. There is a change of some kind that warrants a new marketing strategy, production effort, sales and marketing campaign, and so forth. NNTW products could be completely new to the firm, that is, a new line it had never developed or marketed before, or an extension to an existing line. Some categorizes of NNTW products are:1

  • • new to the firm;
  • • new to a product line;
  • • an enhancement to an existing product; and
  • • a repositioning of an existing product.

For forecasting, the type is not important; the amount of data available for it is the important factor. I will discuss two cases in the next subsection followed by forecasting methods available for each case.

< Prev   CONTENTS   Source   Next >