Optimal Intraproject Learning

Problem definition: Intraproject learning in project scheduling is concerned with the use of learning among the similar tasks in the project to improve the overall performance of the project schedule. Under intraproject learning, knowledge gained from completing some tasks in a project is used to complete similar later tasks in the same project more efficiently. We provide the first model and solution algorithms to tackle this intraproject learning problem.Methodology/results: We model the tradeoff between investing time in learning from completed tasks and achieving reduced durations for subsequent tasks, to minimize the total project cost. We show that this problem is intractable and develop a heuristic that finds near optimal solutions with a strong relaxation that allows some learning from partially completed tasks. Our computational study results identify project characteristics where intraproject learning is most worthwhile. In doing so, they motivate project managers to understand and apply intraproject learning to improve the performance of their projects. A real case is provided by a problem of the Consumer Business Group of Huawei Corporation, for which our model and algorithm provide a greater than 20\% improvement in project duration.Managerial implications: We find consistent evidence that larger projects benefit more from intraproject learning. Our computational studies provide the following insights. First, the benefit from learning varies with the configuration features of the project network, and projects with more complex network possess greater potential benefit from intraproject learning and deserve more attention to learning opportunities; second, noncritical learning activities at an earlier stage are more beneficial and should be learned more extensively; third, tasks that are more similar (or have more similar processes) to later tasks deserve more investment in learning. Learning should also be invested more in tasks that have more successors, where knowledge gained can be used repetitively