Leveraging human learning to develop organizational capacity



Cognitive psychologists have long recognized that individuals learn when performing a task, with this learning leading to reduced performance times and, relatedly, increased capacity. Similarly, the production community has recognized the impact of learning on estimates of per-unit production costs. While predictive models of the effects of learning have been developed, there are few prescriptive planning models that incorporate those predictions.

For an organization where individuals perform different types of tasks, recognizing human learning when scheduling offers the opportunity to direct organizational capacity development at the expense of near-term capacity.  However, solving models that recognize this opportunity presents multiple challenges. First, the quantitative models that predict how experience translates to learning are all nonlinear. As a result, planning models that incorporate these predictions are hard to solve (computationally-speaking). Second, as capacity development is directed in anticipation of customer demands, these models are naturally stochastic.

In this talk, we will present techniques for overcoming both of these challenges. For the first, we will present a technique for representing certain forms of non-linear learning curves with integer variables and linear constraints, removing the need to solve a Nonlinear program. For the second, we will present two Approximate Dynamic Programming-based approaches, with the first only anticipating customer demands in the next period (i.e. a one-step look-ahead procedure) and the second looking further out. We will illustrate the use of these techniques in the context of technician scheduling, and based on an analysis of the solutions prescribed by the approaches, present various managerial insights.