Managers at Work: Beware the Percent Completion Metric

Abstract

Delays in new product introduction can have serious financial consequences (1). It is thus important to estimate accurately the time needed to complete a new product so that marketing promotions and other launch activities are started at the right time. One must also have a tracking mechanism in order that can be forecast early enough to allow deploying alternative business plans. In this article, I take a simple view of project (defined as the difference between the announced and actual product launch); namely, that all are a failure to correctly predict project completion, regardless of the root cause for the delay. Realistically, this is appropriate for R&D projects, where one way of viewing delays is as a failure to account for unanticipated problems in scheduling. These problems, while unanticipated, occur commonly in R&D. Although the specific technical issue is not known in advance, the occurrence of these problems is anticipated and should be incorporated into schedules. My simple view would not be appropriate for some truly rare problems (e.g., the plant burns down). Predicting Completion For discussion purposes, one can consider two ways to predict project completion, although, of course, real predictions will be a mixture of methods. 1. Traditional Schedule Predictions.Often, predictions are based on a bottoms up method, whereby each manager estimates the time to complete his or her subtasks, frequently using Gantt charts. As there can be huge pressures in a company to launch as early as possible, managers are often too optimistic with their predictions. In R&D projects, unanticipated problems and the time required to solve them are often underestimated. Completion dates are often revised when it is clear that the originally forecast date will not be met. The new date is often forecast solely using assumptions about future activities such as, We have proposed solutions for problems A-C. Even if these solutions work, they may introduce new problems. 2. Predictions Based on Past Data.Silverberg suggested a graph of predicted completion dates vs. the date they were made (2). As the project progressed, one could use this graph to assess the likelihood of the current predicted launch date. One value of this approach is that it uses actual performance of past schedule predictions, rather than relying solely on assumptions of future events. Tracking Project Progress Given that a project schedule contains a launch date, a tracking mechanism is required to give management an idea of how reliable the estimated launch date is. One such metric is completion of all tasks, i.e., treating completion of the project as the summation of completion of each task. For most R&D projects, this metric is very misleading and will usually lead to over-optimistic completion predictions. A remedy is suggested which in concept is similar to a learning curve, originally suggested by Duane (3). Note that the completion metric could have other names, such as percent of specifications met or percent of parts released. If task completion times within a project were (approximately) uniformly distributed, then the metric of completion of all tasks would be meaningful. This is because there would be a linear relationship between completion and calendar time. Thus, if 50 of all tasks had been completed in five months, we would predict project completion within 10 months. For almost any R&D project, this assumption of uniform task completion times would be quite unlikely, and hence a poor model. A more realistic model would have short completion times for some tasks, such as documentation, building breadboards or prototyping user interfaces. Much longer completion times would be required for other tasks, such as solving unanticipated technical issues. This relationship could be modeled by an exponential task completion time (Figure 1). …