New faculty members and principal investigators should be wary of applying practices developed for industrial management to managing students, postdocs, and other scientists.
At last count, nearly a quarter of the employed adults who live in my neighborhood were software developers. Another good-sized group includes managers, business development analysts, corporate strategists, and consultants. Although this suburb of a major metropolitan area tends toward affluence, it is not unusual. Its vocational makeup reflects a well-established fact: A large and growing percentage of value creation in the 21st century economy results from "knowledge work."
So casually is the expression "knowledge work" used that its meaning has become vague. In management research, we've tended to define it as work in which value is created primarily through manipulation of ideas or symbols, and which occurs primarily in intellectual domains. By this definition, there's nothing new about knowledge work. It's been the main activity of academic institutions for hundreds (maybe thousands) of years. What's new is the degree to which it has ventured outside academia, into industry and other major institutions.
Knowledge work has always been important in knowledge creation, but in recent history knowledge creation has begun to correlate much more obviously with wealth creation. As a result, the business-minded have increasingly applied to knowledge work practices that were developed to manage industrial work. In general, this has worked rather badly.
Traditional Models of Supervision--A Poor Fit With Knowledge Work
Consider state-of-the-art economic models that underlie traditional notions of supervision, the so-called "agency models." To be fair, not all management scholars approve of these models. Behavioral scientists dislike them particularly. Nevertheless, agency models embody the assumptions most common in the design of modern supervisory arrangements.
Agency models present supervision in caricature, consisting of a "principal" (the employer), an "agent" (the employee), and the relationship between them. The motivations of these actors are simple. The agent dislikes expending effort but likes getting paid. The principal dislikes paying the agent but likes the valuable work that the agent does. The objectives of principal and agent are thus diametrically opposed: The agent wants to be paid as much as possible for doing as little work as possible, whereas the principal wants to get as much work as possible from the agent while paying as little as possible. Contracting is complicated by the fact that the principal can't directly see how much effort the agent is expending but can only observe a result of the agent's work, which also depends on random factors. For example, a salesperson's total sales might depend not only on how hard he works but also on the weather during a given week. The primary conclusion of agency theory is the importance of tying pay to performance, to provide employees with appropriate incentives.
The basic elements and assumptions of traditional supervisory practice are all present in this simple model. You can see a rationale for current trends toward measurement and accountability in everything from research funding to educational testing in public schools. You can see the rationale behind the modern passion for "incentivizing" performance. Yet there are several obvious cracks in the model as it applies to knowledge work:
The difficulties in observing knowledge work are more profound than those assumed by the model. Not only can't a supervisor observe effort directly in knowledge work, sometimes the supervisor can't understand what the worker is doing and may not be qualified to judge results. Because knowledge work occurs in intellectual domains, it is also more difficult to see causality and to attribute results to particular worker actions. Results measures often don't faithfully capture the results you really care about.
"Agent" motivations are inconsistent with those of knowledge workers. Knowledge workers are often interested in their work and motivated by a desire to do it well. Agency models suggest no way of leveraging these worker motivations.
In traditional agency models, effort is a single dimension. The productivity of knowledge work, in contrast, often has to do with how effort is allocated across multiple dimensions. By definition, knowledge work is more about how smart you work and less about how hard you work. Incentive schemes intended to extract more effort from knowledge workers often distort their effort allocations, forcing them to apply effort in the wrong places.
Traditional Structure of Physical Work--A Poor Fit With Knowledge Work
Physical work--the realm from which traditional management theory emerged--is often characterized by high costs of experimentation: of changing work processes and of trying unsuccessful work processes. In building automobiles, for example, you need to make sure you plan in tremendous detail before you actually install equipment or build cars. If you get something wrong, it will be expensive to retool the factory and you'll have scrap costs when you throw away ruined materials. Hence, physical work has often been structured to plan intensively and move only gradually toward implementation. Work proceeds in linear, discrete steps. The overall thrust is toward getting it right the first time.
The problem with this approach is that it does not incorporate change or innovation very rapidly. If you must plan everything in detail before you do it, a late change to a product or process throws a monkey wrench into the gears. Consequently, traditional physical processes freeze process changes at some point in a work project.
Managing Knowledge Workers
The agency model presumes the employer is able to accurately measure employee performance. Yet measuring performance is always difficult, and in knowledge work it is especially difficult. If you have no real chance of observing, understanding, or attributing the results of employee work, you become much more dependent on employees' willingness to openly communicate the meaning of their work. Fortunately, knowledge workers often have a commitment to the work itself that makes them inclined toward information sharing. Strong, agency-theory based incentives typically interfere with open communications by giving knowledge workers reasons to compete with their colleagues and to horde information about how to perform well on official performance measurements. Best practice calls for emphasis on relationships, collaboration, and professionalism, and for de-emphasis of formal performance measures.
The cost structure that drives physical work toward linear, sequential work processes is not inherent in knowledge work. "Retooling" in intellectual domains is often (although not always) much less costly than it is in physical work, and there are fewer "scrap costs." Knowledge work is therefore less constrained than traditional physical work by the need to get it right the first time and can instead be more iterative and more oriented toward exploring, experiencing, trying, and trying again. In knowledge work, rapid experimentation can substitute for detailed planning.
Successful knowledge work processes often iterate frequently (e.g., daily). They are characterized by alternating periods of unstructured work by individuals and small groups and structured "pulling in the reins" by managers to integrate work. Such processes often look messy, even when healthy and productive. Team size needs to be controlled, because the complexity of the "reining in" process can become overwhelming if there are too many people involved. When the process is working well, each iteration introduces new ideas into work processes.
Two main principles of knowledge-worker management can be summarized as follows:
Emphasize collaboration and professionalism; de-emphasize incentive schemes and performance measures. Play up knowledge workers' natural tendencies to be committed to their work and its overall objectives.
Emphasize iterative work structures rather than linear, sequential ones. Don't overplan. Alternate between unstructured individual experiences and structured integration of individual work.
Note that not all knowledge work has the characteristics that make these more modern trends applicable. You have to make sure that your workers are indeed committed to their work before relying on that commitment in collaboration. You must make sure the cost of iteration is in fact low before you structure work iteratively rather than linearly. If the prerequisites are satisfied, though, you can manage knowledge workers in a way that is not only more effective but also more humane and pleasant for all involved. Good research managers understand this implicitly: that relationships based on professionalism and mutual respect work far better than scales of accountability and incentive schemes in most knowledge-work settings.
Robert Austin is Associate Professor at Harvard Business School.
This article first appeared in Science's Next Wave, 26 April 2002. If you're interested in exploring in more detail how Robert Austin's theories on managing knowledge workers might apply in a laboratory setting, read his interview  with Editor Jim Austin.
Comments, suggestions? Please send your feedback to our editor .