The illusory diffusion of innovation: An examination of assimilation gaps

Innovation researchers have known for some time that a new information tech nology may I be widely acquired, but then only sparsely deployed among acqu iring firms. When this happens, the observed pattern of cumulative adoption s will vary depending on which event in the assimilation process (i.e., acq uisition or deployment) is treated as the adoption event. Instead of mirror ing one another, a widening gap-termed here an assimilation gap-will exist between the cumulative adoption curves associated with the alternatively co nceived adoption events. When a pronounced assimilation gap exists, the com mon practice of using cumulative purchases or acquisitions as the basis for diffusion modeling can present an illusory picture of the diffusion proces s-leading to potentially erroneous judgments about the robustness of the di ffusion process already observed, and of the technology's future prospects. Researchers may draw inappropriate theoretical inferences about the forces driving diffusion. Practitioners may commit to a technology based on a bel ief that pervasive adoption is inevitable, when it is not. This study introduces the assimilation gap concept, and develops a general operational measure derived from the difference between the cumulative acqu isition and deployment patterns. It describes how two characteristics-incre asing returns to adoption and knowledge barriers impeding adoption-separate ly and in combination may serve to predispose a technology to exhibit a pro nounced gap. It develops techniques for measuring assimilation gaps, for es tablishing whether two gaps are significantly different from each other, an d for establishing whether a particular gap is absolutely large enough to b e of substantive interest. Finally, it demonstrates these techniques in an analysis of adoption data for three prominent innovations in software proce ss technology-relational database management systems (RDBs), general purpos e fourth generation languages (4GLs), and computer aided software engineeri ng tools (CASE). The analysis confirmed that assimilation gaps can be sensi bly measured, and that their measured size is largely consistent with a pri ori expectations and recent research results. A very pronounced gap was fou nd for CASE, while more moderate-though still significant-gaps were found f or RDBs and 4GLs. These results have the immediate implication that, where the possibility of a substantial assimilation gap exists, the time of deployment should be ca ptured instead of, or in addition to, time of acquisition as the basis for diffusion modeling. More generally, the results suggest that observers be g uarded about concluding, based on sales data, that an innovation is destine d to become widely used. In addition, by providing the ability to analyze a nd compare assimilation gaps, this study provides an analytic foundation fo r future research on why assimilation gaps occur, and what might be done to reduce them.