When Intel announces yet another breakthrough in chip technology, the triumph is as much a testimony to the rapid advances of modern development tools as it is to the skills of the research and development team. Indeed, the exponential performance gains of integrated circuits have fueled dramatic advances in computer simulation and tools for today’s design teams. This progress has now come full circle: Today’s complex chips would be impossible to design and manufacture without the tools that they helped to create. Not surprisingly, companies in many fields have invested billions of dollars, expecting that these innovation tools will lead to huge leaps in performance, reduce costs and somehow foster innovation.
But tools, no matter how advanced, do not automatically confer such benefits. In the excitement of imagining how much improvement is possible, companies can easily forget that these artifacts don’t create products and services all by themselves. People, processes and tools are jointly responsible for innovation and development. In fact, when incorrectly integrated into an organization (or not integrated at all), new tools can actually inhibit performance, increase costs and cause innovation to founder.1 In a nutshell, tools are only as effective as the people and organizations using them.
Many new products or services depend on innovations in development tools. My research has found that new tools can significantly increase developers’ problem-solving capacity as well as their productivity, enabling them to address categories of problems that would otherwise be impossible to tackle. This is particularly true in the pharmaceutical, aerospace, semiconductor and automotive industries, among others. Furthermore, state-of-the-art tools can enhance the communication and interaction among communities of developers, even those who are “distributed” in time and space. In short, new tools (particularly those that exploit information technology) do hold the promise of faster, better, cheaper.
But that potential should be tempered: New tools must first be integrated into a system that is already in place. Specifically, they must be integrated into the work that needs to be done, not unilaterally pasted onto existing routines or substituted for what is presumed to be an equivalent.
1. A detailed discussion is contained in S.H. Thomke, “Experimentation Matters: Unlocking the Potential of New Technologies for Innovation” (Boston: Harvard Business School Press, 2003), which is the source of some of the material included in this article.
2. The study methods and findings were published in K.B. Clark and T. Fujimoto, “Product Development Performance: Strategy, Organization, and Management in the World Auto Industry” (Boston: Harvard Business School Press, 1991).
3. The general findings from the second round of research can be found in D. Ellison, K. Clark, T. Fujimoto and Y. Hyun, “Product Development Performance in the Auto Industry: 1990s Update,” working paper 95-066, Harvard Business School, Boston, 1995.
4. Total engineering hours are hours spent directly on projects by engineers, technicians and other employees. Measured activities include concept generation, product planning and product engineering carried out in-house or sub-contracted to engineering firms. The numbers exclude suppliers’ engineering hours, general overhead, new engines and transmission development, process engineering and pilot production. To account for project complexity, the following variables were measured: (1) number of body types per project (for example, two- or four-door sedans), (2) total percentage value of new parts that were designed (platform-type projects typically had values of more than 80%), (3) product category (micro, compact, mid-size and luxury) and (4) the supplier contribution to design. These variables were similar to project controls used in K.B. Clark and T. Fujimoto’s original 1991 study (see reference 2). A regression analysis showed that body type, new part design and product category were very significant (at less than 5%), whereas supplier design contribution had significance (at 12%). The variables’ regression coefficients were used to predict engineering hours for each project (given its complexity), which were then subtracted from the actual value reported by firms. Positive residual values indicated worse than expected performance and vice versa.
5. Total development time is the longest time-to-market measure, extending from the initiation of concept development to market introduction. Other measures that are often used in industry journals measure the time from program or design approval to start of production, which is much shorter and was also measured as part of our study. As with engineering hours, a regression analysis was used to determine the effect of project complexity on development time. The variables new part design, suppliers’ design contribution and product category were significant (at less than 10%); body type was also significant (at 18%). The variables’ regression coefficients were used to predict development time for each project (given its complexity), which was then subtracted from the actual value reported by firms. Positive residual values indicated worse than expected time and vice versa.
6. For a discussion of this assumed trade-off and actual empirical evidence, see K.B Clark and T. Fujimoto, “Product Development Performance” (1991); F.P. Brooks Jr., “The Mythical Man-Month: Essays on Software Engineering,” Anniversary Edition (Boston: Addison Wesley, 1995); and G.P. Pisano, “Development Factory: Unlocking the Potential of Process Innovation” (Boston: Harvard Business School Press, 1996).
7. The data shown were collected for six different subsystems of a car (body-in-white, interior, instrument panel, seats, suspension and engine/transmission) in 18 projects. For simplicity, the data are shown here in aggregated form by reporting only averages. Similar data were also collected for each firm’s supplier base, which showed similar regional differences but also a lower level of tool use by suppliers when compared to auto firms.
8. Detailed prescriptive advice for implementing development tools is contained in chapters 5 and 6 of S.H. Thomke, “Experimentation Matters” (2003).
9. The following discussion draws extensively from S. Thomke and A. Nimgade, “BMW AG: The Digital Auto Project (A),” Harvard Business School case no. 699-044 (Boston: Harvard Business School Publishing, 1998).
10. Physical prototypes in the automotive industry are usually made from metal or other material that allows for functional evaluations. In contrast, partial or full-scale models made from clay, foam, wood or other similar materials are not reported here. In our research, we expected the number of physical prototypes to be affected by project complexity, but the results of a regression analysis showed otherwise. The number of body types per project, the ratio of new part design, suppliers’ contribution to design and product category had no effect on the number of prototypes built per project, and thus it was unnecessary to make adjustment to the reported data (the significance of the regression analysis was greater than 50%). We did observe, however, that car programs with higher expected sales volumes also ended up with larger prototyping budgets that, in turn, led to more of them being built.
11. For details on how the first study data were collected, see K.B Clark and T. Fujimoto, “Product Development Performance” (1991), p. 369.