² The scorecard evolved over time as better metrics were identified.

³ The model is a large system of coupled nonlinear differential equations. Complete documentation is available from the authors upon request.

⁴ The economic growth index was not significant, apparently because the variance in annual demand growth is not great enough to estimate the relationship. Nevertheless, since the macroeconomy must influence the demand for ADI's products, we retain g^e in the model with an income elasticity of unity, which is not significantly different from the estimated value.

⁵ In 1991 ADI acquired one of its competitors, PMI. The acquisition is not incorporated in the model, so comparisons after 1991 are not meaningful.

⁶ Actually these activities took time and were distributed over a several month period. The assumed starting time represents a good midpoint for the startup process.

⁷ It might be argued that the decline in prices was due entirely to market pressures outside of Analog's control. If so, the drop in profits was not a side effect of TQM but the inevitable result of increased competition. Clearly competition has intensified. In the mid 1980s competitors began to produce `pin compatible' substitutes offering identical functionality. ADI's business also began to shift from defense-related applications towards consumer markets where competition is more intense and margins are lower. If ADI used markup pricing exclusively and ignored competitive pressures, the 15.8% decline in unit production costs from 1985 to 1989 would yield the same fractional reduction in average selling prices. Average prices actually fell 17.2%, implying intensifying competition and shifting product mix caused prices to fall 9% beyond the decline in unit costs. If prices had fallen only as much as unit costs, operating income per unit would still have fallen 35% because of the smaller drop in indirect costs per unit.

⁸ Schneiderman left Analog at the end of 1992 and became an independent consultant.

⁹ More sophisticated tools for improvement such as Taguchi and other experimental design techniques can accommodate nonlinearities and interactions. However, we are not aware of any applications of these design methods to managerial processes.