Learning Rate Implications

January 16, 2010 - Tags:

A classic problem studied by researchers from many fields is how firms allocate resources to the exploration of new possibilities versus the exploitation of known certainties. The returns of exploration are more long term, uncertain, and therefore risky. As March22 puts it: what is good in the long term is not always good in the short term. What is good for one part of an organization is not always good for another part of the organization (or the whole organization), and what is good for an organization is not always good for the society. “As organizations learn from experience, this distribution of consequences across time and space affects the lessons learned,” says March. If distances are smaller, feedback is quicker. Thus, experiments that are more local and about the near term provide feedback quickly and tend to reinforce the “local” aspect of learning. Because of these differences, organizations that learn through feedback gained from experimentation and sharing knowledge of outcomes (or adaptive processes) tend to improve exploitation rather than exploration. As organizations specialize and become more and more competent at exploitation, they tend to stay with what they do best. Thus, organizations might gain competence in inferior activities at the expense of not switching to a superior activity. This effect is passed on to other firms with whom the firm interacts. There is also the effect of excessive specialization. Specifically, in the models proposed by researchers, agents are “trapped by immediate positive feedback from competence within a rather narrow domain.” For these reasons, March posits that organizations might want to control learning, and he suggests some ways of doing so.

For example, slow learning might preserve sufficient diversity among employees, thereby preserving exploration until convergence of ideas occur. Slow learning also avoids false association of causes to events (in the theory of learning, that is called “superstitious learning for obvious reasons”). In many places, we see the emphasis on slow learning in Toyota’s supply chain. March also suggests that a modest amount of turnover preserves the heterogeneity until new employees are socialized into the organization and provided with the impetus for exploration. Too little turnover leads to greater homogeneity and less deviation from the “norm,” whereas substantial turnover dissipates learning. Toyota provides “turnover” by rotating its employees through a variety of tasks.

Even though a modest amount of turnover is good, rapid socialization reduces the impact of new thinkers on the organization. Therefore, employees should be brought up to speed slowly. At Toyota, even suppliers and dealers are brought up to speed and attain full partnership slowly. Simon writes that “tasks of management are quite different in organizations that can recruit employees who are pre-fashioned, so to speak, than they are in organizations that wish to create and maintain, along some dimensions, idiosyncratic subcultures.” If the idiosyncratic aspect we wish to create is “systems thinking,” then Toyota’s management has a formidable task. If it is protecting the bastion, then bureaucratic training might be necessary. If anyone can step in and do the job, then a mass production approach with limited on-the-job training might suffice.

Nevertheless, Toyota faces the risk of excessive exploitation by its supply chain partners and thus has to inculcate similar thinking into each of its partners. That fact might explain the “vague” instructions given to its partners. For example, Toyota might ask a supplier to “explore the range of improvement possible.” Let us say the possible improvement is a 20 to 30 percent reduction in weight. The supplier might come back and present what it has learned about the design and what can be achieved. The target is then gradually narrowed down. As a general manager has said: “This process allows the (user) to understand trade-offs and set targets to produce the best possible design.”24 In universities, we train our Ph.D. students in a similar fashion. We give them vague targets, such as, “See what this assumption does” or “Can you relax this?” or even, “You may have forgotten something.” Many times, the student comes back and suggests, “You said, ‘Try doing this,’ but I found something else”—we are looking for such opportunities to learn.

Herbert Simon elaborates that in some cases research ideas get constrained by the market and customers; in other cases the needs of the customers are well known and the flow of ideas is in the opposite direction. In the former case, research can be facilitated by setting goals that have an element of exploration in them and by getting feedback on the results, both anticipated and otherwise. Simon acknowledges that for such transfer of ideas to occur, disparate groups need to respect one another’s skills, understand the others’ problems, and actually have experienced, in sufficient numbers, the other groups’ activities and processes. For example, in a recent client engagement, one of the authors had the opportunity to work with a team member who had worked for several years with the client in many of the client’s businesses. The person could be reliably counted upon to draw upon his experience and write down a business process in 85 percent of its detail. We realized that the remaining 15 percent could be had only by observing the processes at work! People were using rules written and unwritten to make decisions. Only by asking them for daily feedback based on our replication of their work were we able to uncover another 14 percent of the rules. The last 1 percent still eludes us.