by CIO Staff

GM NA’s IT Project-Tracking Dashboard: Give Regression a Try

Oct 01, 20012 mins
Project Management Tools

GM North America’s dashboard tool takes an intuitive approach to summarizing large amounts of project status data with a color scheme that everyone understands. I’m sure it goes a long way toward helping upper management keep an eye on projects underway.

The one thing I’d caution is that GM should make sure that the right data is being summarized and represented. Project tracking is, after all, a forecasting problem?the goal is to reliably point out projects that need help (that is, projects whose outlook is poor without help) without bringing undue attention to projects that are getting along just fine. I’m not certain that the four metrics included in GM’s tool forecast problems reliably. Evaluations of business results and risk, in particular, are made fairly subjectively. As a test of the metrics’ predictive power, GM North America might look at historical data for all projects that were consistently coded green in the risk category, for example, and see if more of them were actually successful implementations than those where risk was coded red. If not, the data being summarized isn’t of much use for helping manage project success.

An “actuarial” analysis of GM’s extensive project history might also be useful. A formula could be derived that considers only objective criteria and computes probabilities of potential disasters. Factors such as changes in project management, skill levels of staff, level of sponsorship and number of business units serviced could all be put into a regression model to see if they are correlated to future undesirable outcomes, such as cancellation of the project. This is a practical analysis that has already been implemented at other companies.

Furthermore, with an objective model of risks, GM’s Process group wouldn’t need its system of assigning arbitrary point values to the colors for its 12-month views. The color codes are already somewhat arbitrary and subjective, but assigning numbers to them would, in the field of decision theory, be considered an “information destroying” step. That’s because the forecasting ability of this numerical score is probably even worse than the color assignments themselves. A regression method would certainly produce better forecasts. I’m also certain that the size of GM North America’s project portfolio would easily justify the effort to create a more statistically sound approach to project tracking.