Given the current state of knowledge, there are at least two ways in which public policy could support workplace innovation. One is to encourage research on the conditions in which innovative work systems flourish and generate more positive-sum arrangements between management and labour. Another is to foster institutional developments that would incite employers and unions to work on a social compromise which would be more attuned to recent and ongoing developments in the sphere of production. Each of these two propositions deserves some elaboration. The first, on directions for research, will be developed briefly here, while the notion of institutional support will be discussed in the Conclusion.

We would suggest that progress in research on innovative work systems lies very much in finding the middle ground between two extreme views. At one extreme, some literature suggests there exists something of a "one best way" in terms of organizational characteristics which, if they could be found and properly tested in a given firm, could be replicated by competitors and even have universal applicability. According to this view, innovative work systems would have public goods characteristics and only give a short-term competitive advantage, a factor which would deter investment in organizational development. While not so many specialists on workplace relations explicitly hold such a view about the possibility of a "one best way," it certainly underlies many management fads as well as some highly influential books on production models (for example, Womack, Jones and Roos, 1990). Indeed, in their synthesis on work systems and firm performance, Becker and Huselid observe: "A related debate that runs throughout this literature is whether there is a best HRM system with universal applicability, or whether the strategic impact of HRM is contingent on the fit between the HRM system and corporate strategy" (1998: 58).

Considering the specific features of any productive organization and the complexity of the innovation process, this notion of a "one best way" does not fit with reality. But the other extreme, held by many experts in organizational research, is hardly more fruitful. Insisting too much on the specificity and idiosyncrasies of every workplace and stressing variations from case to case often leads to the view that optimal organizational choices are "totally contingent." In other words, they see so much firm specificity that no pattern can be inferred. On this account, there would be only limited prospects for theory and public policy. Our point is that each of these two extreme views leads to an impasse. Workplace research suggests, however, that there may be a middle ground, to be outlined briefly here.

The way forward is to examine the conditions under which mutually reinforcing practices happen to fit with a given technology and system of production, and hence contribute to organizational performance. There is some emphasis in the literature on the fact that only clusters of complementary practices can make a difference. In the words of Ichniowski et al., "there are no one or two 'magic bullets' that are the work practices that will stimulate worker and business performance" (1996: 322). Some of the studies reviewed above go further in indicating the set of work organization and HRM practices that adequately support a given technology. We have in mind MacDuffie's research in the automobile industry (1995) and the comparative studies of Ichniowski et al. (1997, 1999) on steel finishing lines. In the latter case, a given set of practices produced similar positive results for productivity and quality in the United States and Japan.

Our own observation of different types of batch production and of continuous process technology indicates that systems of production really make a difference as regards the possibilities for work reorganization. For instance, we studied the progress of teamwork in two aluminium smelters. In such a continuous process, also in operation for example in steel and other metal smelters and in oil refineries, production workers have to monitor technical systems. They have large areas to cover, and performance does not depend directly on individual effort on a day-to-day basis. Rather it requires good co-ordination within and between teams, as well as with employees (e.g. maintenance craftsmen, engineers and technicians) whose task is to assist production. Such a system of production creates favourable conditions for self-regulation and limited direct supervision. In contrast, in routine assembly work, where performance depends directly on the volume and quality of output of each individual on a daily basis, it is more difficult to do away with direct supervision. Some research suggests that teamwork may then lead to peer pressure and excessive forms of social control (Barker, 1993; Sewell, 1998).

Of course, our aim is not to suggest a form of technological determinism but rather to remind that technology matters. In fact, the key factor is not so much technology per se as the system of production more generally. It may be worth re-examining Woodward's early research (1965) on the links between production systems and organizational structure. In short, she combined the 11 categories of production systems observed in British manufacturing into three main groups, namely unit and small batch production, large batch and mass production, and process production. Her theory holds that unit and small batch production leave more uncertainty to be managed by the social system. Recent observation of small batch production shows that work is hardly rationalized and standardized by technology and that frequent adjustments (different product specifications, new set-ups, etc) have to be made to production lines. This tends to demand more direct supervision. The contrast between this and a modern aluminium smelter, which is at the other end of the continuum, is striking. Conditions in the latter are favourable to the operation of teams with very little supervision. The hypothesis could be made that as we move from small batch to continuous process, production is more capital-intensive but can be operated efficiently with less supervision. More empirical research is also needed to consider related aspects of the system of production which have consequences on patterns of management control.

These propositions need to be validated through further empirical research; but they also indicate that research has made some progress in identifying the conditions under which a particular set of working practices may find relatively fertile ground.