Perhaps the most critical use of diffusion models is to forecast first-purchase sales volume. The value of predicting sales or adoptions cannot be overemphasized. For example, errors in sales projections can trigger a series of adverse reactions affecting budget forecasts, operating expenses, cash flows, inventory levels, pricing, advertising outlays, and so on. This is critical when new products are launched because typically there is little relevant information to rely upon. As important as forecasting is however, most executives will tell you that any given new product forecast of sales or adoptions will be wrong (Kahn, 2002).
For new products the likelihood of large forecasting errors is typical and there are several factors that increase both the probability and the scale of error. One factor is the forecasting of high-tech products. With high-tech products change is frequently rapid and the rate of new product introduction is fast therefore data series are often short or non-existent. Also, high-tech innovations are often more radically new, making their adoption more important. Therefore a different set of new product forecasting models may be more appropriate than when forecasting sales of low-tech products (Lynn, Schnaars, & Skov, 1999).
But what classifies a product as high-tech? Shaklin and Ryans (1984) suggest the following criteria for high-tech products: a strong scientific basis, new technologies that can quickly make existing technologies obsolete and new technologies whose application create new markets and demand. Based on these dimensions, products with no technical basis (e.g., food), products where improvements are gradual and non-threatening to the core technology (e.g., washers/dryers), and products whose market is primarily replacement sales (e.g., televisions) can be eliminated. Further research characterizes high-tech products along two additional dimensions: uncertainty and switching costs. High-tech products are characterized by higher levels of uncertainty as a result of both the rapid rate of technological change (Norton and Bass, 1992 and Heide and Weiss, 1995) and the lack of relevant prior experience by adopters (Von Hippel, 1986 and Von Hippel, 1988). Switching costs may prohibit potential adopters from purchasing a high-tech product due to earlier commitments to existing high-tech products (Moriarty and Kosnik, 1989 and Heide and Weiss, 1995). A more straightforward way of classifying high-tech products that is consistent with previous research (Lynn et al., 1999) is to use SIC codes to select a high-tech industry.
A second factor that increases both the probability and the scale of error is forecasting organizational adoption. Traditional models of organizational buying behavior include individual characteristics, interpersonal factors and organizational factors as important variables affecting the organizational buying process. Marketing variables such as mass media promotions and word-of-mouth may impact diffusion at both the consumer and organizational levels but organizational diffusion literature suggests additional key variables that may require separate models (Fern & Brown, 1984). Specifically, Framback and Schillewaert (2002) identified adaptor characteristics such as firm size, structure and organization innovativeness or strategic posture as factors that influence the acceptance of new products by organizations.
An additional consideration in organizational diffusion is that while consumer diffusion research assumes there is only one purchase of an adopting unit, organizations rarely buy a single unit since purchases are usually for a group of users. As such, a single “adoption” of a new product by an organizational buyer may include thousands of units and have a great effect on its ultimate diffusion. Lastly, while most consumer diffusion research assumes an innovation is adopted by all potential users, organizational diffusion research suggests that investment costs, network externalities, and competitive pressure can all trigger the rejection of an innovation even if the adopters' preferences are favorable (McDade et al., 2002). In summary, high-tech products may not always diffuse among a population of organizations according to theories developed in the marketing literature based on consumer durables.
Over the years, modeling research has raised questions about the forecasting accuracy of macro-level diffusion models (Bernhardt and MacKensie, 1972, Collopy et al., 1994, Heeler and Hustad, 1980, Meade and Islam, 2001, Lynn et al., 1999 and Rao, 1985). For example, Mahajan, Muller, and Bass (1990) point out that more empirical work is needed to identify conditions under which macro-level diffusion models work or do not work. Likewise, Collopy et al (1994) suggest that despite the large research base on macro-level diffusion models, only a handful of studies has empirically examined the conditions under which they are most appropriate (Meade, 1984 and Rao, 1985).
This study looks to examine the forecasting accuracy when applying macro-level diffusion models to high-tech product innovations among organizational adopters. Additionally it explores whether the accuracy of macro-level diffusion models differs according to the impact of the new product. Exploration is done in two ways, first by validating six widely used forecasting models for 39 high-tech product time series over 3 ex ante forecasting periods, and second by comparing the accuracy of the forecasting models for high-tech product innovations of both incremental and radical impact. The purpose of the study is to assist forecasters in choosing appropriate models for predicting organizational adoption of high-tech products once the product is on the market and initial sales data becomes available. This forecasting situation is particularly relevant because research has found that relative to consumer firms, sales-based forecasting is widely used by industrial firms (Kahn, 2002).
