The extent of authority and autonomy granted to the project manager and to the individual team members has a significant impact on the ability of the project team to meet stakeholders' goals and expectations. This is particularly true in dynamic environments, where resources, requirements, and other conditions change rapidly, and a great deal of innovation is required. In this paper we provide some recommendations regarding the structure and flexibility of project teams operating in turbulent environments. The recommendations are based on a set of experiments carried out with agent-based simulation, a novel tool for organizational design. The observed results show that under high uncertainty, loose designs for project teams outperform tight organizational structures.
Today's project managers and the executives responsible for the organizational project portfolio face a double challenge, especially in research and technology development fields. On one hand, the environment in which they operate is becoming more dynamic and uncertain: new products, technologies and capabilities become available (and, after a relatively short time, become obsolete) at an increasing rate; suppliers and competitors join and leave the market on short notice; and customer requirements and expectations change and evolve on even shorter lead times. On the other hand, the same environment dynamics create tremendous pressure to deliver customer and stakeholder satisfaction, including market and financial success, without which the firms that execute the projects will not survive.
This type of turbulent environment raises several pressing management issues. Consider for instance the research division of a typical corporation in the telecommunications, technology or pharmaceuticals fields. The division employs mainly scientists and engineers, many with advanced degrees, and utilizes costly research infrastructures. Some of the dilemmas facing the executives and managers are: should the highly trained (and consequently costly) people be allowed to form project teams on their own? Will they be allowed to elect their own leaders? To what extent will they be able to join and leave project teams based on their professional interests? In the absence of a well-defined paying customer, how much leeway should the project teams be given in defining their project objectives and scope? How often should these change? Who is authorized to make commitments to external stakeholders? What happens to these commitments when the individual team members or stakeholders change? In summary, what is the optimal amount of autonomy at the individual and team level that will yield the best contribution to the firm? These are key issues for the design of project teams and for the way projects are to be selected and managed.
The effects of task uncertainty and organizational structure on performance have been studied for quite some time, mainly under the paradigm of contingency theory (see for instance Burton & Obel, 1998), which argues that the best way to organize is contingent upon a variety of factors, including environmental characteristics such as those mentioned above. A general statement drawn from contingency theory is that a centralized organizational structure is more effective under low uncertainty conditions, while a decentralized structure is preferable under high uncertainty. However, empirical studies such as those cited by Kim and Burton (2002) provide mixed results that require further research.
In this paper we examine the performance of project teams in turbulent environments that are characterized by high uncertainty regarding the scope of the project and combined with uncertainty and ambiguity regarding the composition and structure of the teams themselves within the organization. The paper is organized as follows. In the next section we provide a brief review of the methodological tool that was applied in this study – computational simulation. Then we describe the model of the organization and its environment that was the basis for the simulation experiment, and the specifics of the experiment that was carried out. The following section presents the results of the simulation and the findings of the experiment. We conclude with some practical recommendations and suggestions for further research.
Recently, the use of computational techniques in organizational research has been accepted as a complementary approach to experiments and field studies (Kaplan & Carley, 1998). The benefits of computational simulation are multiple. First, simulations are inherently dynamic, so they provide a powerful way to study the complex organization dynamics, which are almost impossible to observe using static models. Hence, simulations contribute to developing more realistic models. Second, simulations present the unique characteristic of enabling the control and manipulation of all problem variables, facilitating the observation of all possible combinations. Third, simulations accelerate the research because large sets of scenarios can be simulated in very short periods of time. Finally, simulations enable the testing and the developing of new theories, which would be very difficult using case studies or field studies.
The use of simulations in organizational research is still limited in extent but very important in content. Levinthal and March (1981) showed the implications of environmental uncertainty on the learning and innovation processes of organizations. Lounama and March (1987) used simulations to study the dynamics of cooperative learning. Lant and Mezias (1990) introduced a taxonomy of three types of organizations characterized by their search strategies: fixed type (organizations that do not evolve), adaptive type (organizations that modify their search strategies when the environment changes), and imitative type (organizations that copy the search strategy of the leader). They discovered that under low uncertainty, adaptive organizations outperform the other types. Under high uncertainty, adaptive and imitative organizations perform better than fixed ones, and imitative organizations usually have troubles in the long run caused by the follow-the-leader trap. The same authors observed that the increase in uncertainty generates an increase in the changes made in the organizations and modifies the selection processes (Lant & Mezias, 1992). Mezias and Lant (1994) studied the survivability of imitative and fixed organizations in competition. Mezias and Glynn (1993) found that the assignment of more resources to search in the context of routine functioning increases the level of refinement of current technologies, rather than the level of innovation. March (1991) studied the implications of adaptive learning in the choice between exploitation of current activities and exploration of new activities. The paper argued that adaptive learning creates preferences for exploitation rather than exploration. Carley (1992) used simulations to study the personnel turnover in learning processes. She found that teams learn better and faster than hierarchies, and that hierarchies are less affected by personnel turnover. Kaplan and Carley (1998) introduced COMIT, a prototype designed to model both macro and micro behavior of very small organizations. Kim and Burton (2002) carried out a simulation study on the relationship between task uncertainty, level of centralization and project team performance using SimVision.
In our view, organizations are complex, dynamic, nonlinear, adaptive, and evolving systems. Consequently, they can be simulated by using collaborating agents dealing with uncertainty. Task uncertainty affects team performance negatively, but the degree of negative effect depends on the level of centralization of the team decision-making structure. As task uncertainty increases, decentralized teams suffer less in terms of project duration and costs, but more in terms of quality. Agents, artificial and human, display adaptive behaviors that are difficult to predict by analytical methods. Computational analysis could be a valuable tool for the study of these behaviors, to prove concepts, or to determine the consistency of a theory. In this research we used an agent-based simulation to study the impact of structural flexibility on project team performance when the environment is turbulent.
This research follows the Computational Organizational Theory approach (Prietula, Carley & Gasser, 1998) viewing organizations as computational entities capable of processing information. Internally, organizations are complex information processing systems composed of multiple distributed agents in cooperation. These agents are assigned tasks, resources, and responsibilities. Agents accomplish their tasks processing information and requesting the collaboration of other agents. The interaction among the agents results in external organizational properties that can be measured and analyzed.
For the purposes of this study we modeled the organization as consisting of a number of individuals of three hierarchical classes: top management, project managers, and project team members. Each class in the hierarchy has different characteristics that determine their behavior towards other individuals and towards the environment. The relevant characteristics and their respective values are described later on. Initially the individuals belonging to the different levels of the hierarchy are not organized in any particular way. They create and join project teams and higher-order structures as the simulation progresses.
The organization exists in an environment in which opportunities for value-creation present themselves randomly and their location in the environment is unknown to the individuals. Moreover, their location may change randomly across the environment over time. The rates of occurrence and of change of these opportunities reflect the turbulence of the environment. We will refer to these opportunities as tasks. The objective of individuals at all levels is to identify and perform as many tasks as possible, in order to accumulate as much value as possible.
In the computational simulation, each individual in the organization is represented by a software agent, capable of simulating its behavior in terms of initiative in identifying tasks for execution, following instructions from higher-level agents, and competing with other project teams. These software agents exist in a software program that simulates their combined behavior and is capable of showing their location and trajectory in the two-dimensional environment that contains the task opportunities. The simulation was constructed using object-oriented programming.
Agents move randomly at a fixed speed. Each agent has a maximum detection range (DR). Any task or agent within the detection range is automatically detected. Because the environment is dynamic, as soon as an object (another agent or task) moves outside the detection range, it becomes unknown for the agent. We will refer to the three classes of agents (top management, project managers, and project team members) as Alphas, Betas, and Gammas respectively. For the purpose of our experiment, agents have been defined so that they copy the movements of any higher-class agent within the detection range. Thus, the Gammas will copy any Beta or Alpha in their vicinity; the Betas will follow the Alphas; and the Alphas remain independent. This “copy behavior” generates the opportunities for the creation of teams that share common goals and direction.
The actual experiment consisted of simulating the behavior of a group of agents in a given environment over a period of time. We used a set of 20 agents (three Alphas, six Betas, and 11 Gammas) randomly distributed in a two-dimensional space of 400 by 400 pixels, and with randomly assigned motion courses.
There were two factors in the design of the experiment. The first factor represented the degree of autonomy or flexibility of the agents, and was implemented by means of a binary state variable named “avoid collisions” (AC). When AC is set to True, the agents will change their course maneuvering to avoid the collisions according to the minimal range allowed as specified in a parameter called “avoid collision range” (AR). If AC is set to False then the agents will not change their course. The two different values of AC correspond to two different organization models. If AC is set to False, then collisions (distances = zero) occur and a tight organizational structure appears. If it is set to True, then a more flexible pattern appears. We called these two organizational models “tight structure” (TS) and “loose structure” (LS) respectively.
TS is characterized by an absolute order. This state represents the mechanistic type of organizations. All the agents follow the prevailing Alpha and there is no room for improvisation or innovation. In LS, the organization is flexible enough to accept changes in the environment (e.g., changes in the location of the tasks available for processing) without collapsing, and yet it has enough structure to cooperate (e.g., process the tasks). As the simulation runs, the agents start to find leaders (Alphas or Betas) and some grouping pattern appears. This state models the organic type of organizations.
The second factor in the experiment represented the extent of uncertainty or turbulence of the environment in which the agents move looking for tasks to execute. It was implemented by means of a global binary variable that indicated whether the tasks remained static in the environment or were allowed to move randomly at a specified speed (SP). These two states corresponded to two levels of uncertainty in the environment, which we refer to as “static” and “turbulent”.
The simulation model was fine-tuned by setting the detection range (DR) to 10 pixels, the avoid collision range (AR) to 2 pixels, and the speed of the tasks in the environment (SP) to 5 pixels/unit of time. We found that these values resulted in a good differentiation between the TS and LS models. Each of the different combinations of structure and environment was simulated over a period of 5,000 time units. We measured the following outcome variables:
- 1) The number of agents who became organized and joined a team (i.e., found another agent from a higher class whose behavior they could copy or at least one agent from a lower class who would copy their own behavior). This is a measure of the extent of task-oriented organization that was achieved among the original set of individuals.
- 2) The number of agents who became engaged in processing tasks. This is a measure of the success in dealing productively with the environment and taking advantage of the opportunities that were present.
- 3) The cumulative work (CW), computed as the sum of the number of agents engaged in processing tasks multiplied by the time interval, for all the intervals.
Exhibit 1 shows the number of organized agents over time in a static environment (i.e., tasks do not change over time). We notice that the behavior over time is almost the same for both the TS and the LS models, with TS reaching its maximum a little bit later and exhibiting a little bit more variability than LS.
The advantage of LS versus TS becomes much clearer when we consider the number of processing agents, which is the number pf project team members actively engaged in executing tasks that create value. The evolution of this measure over time for the static environment is shown in Exhibit 2: for LS it reaches the maximum after 1500 units of simulated time, while for TS 3500 units of simulated time are required to reach its maximum, which is at a lower or the same level. If we consider the cumulative work (CW) the superiority of LS over TS becomes clear as shown in Exhibit 3. Value-adding work started to accumulate almost immediately for LS, while it took over 1500 time units to start in the TS model. At the end of the simulation, the LS model accumulated about 2.5 more work than the TS model.
In the turbulent environment, the number of organized agents was not recognizably different in the two organizational models. The number of processing agents was higher over time for the LS model than for TS model. However, the advantage of LS over TS is clearly seen in the cumulative work (CW) curve (Exhibit 4), which grows at a faster rate in the LS model. Observe that the level of 2000 units of CW is reached at t=1550 with the LS model. Using TS, it requires 4700 units of time to reach a similar value.
Another interesting observation from the simulation was than in the TS model, after a certain time, only one Alpha was able to gather a team and prevail, while the others remained alone. In this state the processing capacity of the grouped agents is maximized. However, the structure degrades the joint detection range. Consequently, this design is less effective in finding new tasks. Specially, if the environment changes drastically, then the ability to find and execute value-adding tasks declines significantly.
The LS model is the best fitted to cope with turbulent environments as represented by the random location of tasks. Another interesting behavior is also observed in this state: when two or more teams approach too much, some agents react by changing from one group to another. This side effect of the simulation is a close metaphor for the social dilemma of collaboration vs. defection that was described by Huberman and Glance (1998).
In the above experiments, the tasks changed their location randomly every 100 simulated time units. To test different levels of volatility, we conducted experiments at higher turbulence rates, with tasks changing every 50 and 10 time units. In all cases, LS design showed the best performance in terms of cumulative work accomplished (CW). Our experiments agree with the conclusions of (Kim & Burton, 2002) concerning highly turbulent environments. However, for the extremely turbulent case, we did not found evidence that a centralized team (tight structure) could perform better than a flexible one.
The observed results show that in turbulent scenarios, where the tasks are volatile and a great deal of innovation is required, artificial societies that exhibit flexibility (i.e., the LS organizational model) prevail. Too much structure or too little structure contribute to the extinction of the society. This finding is consistent with those of Brown and Eisenhardt (1998) regarding the traps of chaos and bureaucracy, and also with those of Lant and Mezias (1990) who found that under uncertainty, adaptive organizations outperform the other types.
Conclusions and Further Work
The results of this simulation study point in a clear direction. In uncertain environments, when the goals and tasks are not fully defined, it is best to avoid a highly structured and tightly controlled project team structure. Maintaining a loose organizational structure with significant flexibility is even more important in dynamic situations where tasks are not only ill-defined, but also change and evolve over time.
By allowing the professional staff to form their own teams, join the leaders that seem most appropriate, and even change teams as their interest evolve, the overall capability of the organization to identify and execute value-adding tasks is maximized.
In turbulent environments, a tight team structure not only accumulates less value, but also inhibits the development of management and leadership talent, which may have an indirect impact on the ability to attract and retain high-potential individuals.
This experiment with simulated societies allowed the observation of behaviors that otherwise would be very difficult and time consuming to study. The experiment also shows the benefits of computational analysis as a research tool in the organizational theory field. Another issue of interest that could also be studied with agent-based computational simulation include the impact of organization structure and of adherence to routines and standards on organizational learning, innovation and process improvement that lead to enhanced performance.