History, Estimates and PERT
By Jon M Quigley
Schedule Failures Due to Poor Estimates
Of the numerous project failure I have experienced or witness, time and schedule is one of the more frequent occurring. Sometimes we may see a schedule that is borne out of an executive’s fancy and not reality (sometimes it is marketing). Sometimes we are squeezed to get a project out as pressure for other reasons. Time constraints are not bad, it is a way to ensure we are being good stewards of the company’s resources. This pressure can be good as it can be a catalyst for creativity and generating solutions. That is the subject for later posts.
Duration Estimation and Historical Information
When planning the project schedule it is often practical to look at our historical record for estimating. Our historical record, if generated over sufficient time, provides us with a glimpse of the capabilities of our Project Management Organization as well as various line functions. A single point source of historical record (for example one project) does not provide even a glimpse of the envelope of possible outcomes or duration.
Even when we have abundant historical data, we are not out of the woods. The measurements have limitations. There can be questions around the validity or veracity of the measurement that may be misleading. In companies that have a significant political component, there may be some exceptions not noted to make the numbers appear “good”. There may be changes in the team members and certainly there will be changes in project dynamics and scope details. All projects are unique, even if there are elements that may be common. Therefore historical records, like all estimates, are still subject to variation, and with enough representative historical data we begin to understand the variation.
Estimating via PERT
Another estimating technique is called PERT. PERT attempts to fit a set of estimates to a normal distribution curve. By the way, the assumption that the range of distribution duration for a certain task fits a normal distribution may not be valid. It does, however, provide a range of possibilities based upon an estimation of three points. Those three points are, most probable, pessimistic, and optimistic. Experience suggests to me that one point estimate durations tend to be on the optimistic side, so asking for a range helps mitigate this optimism. The equation weighs the most likely heavier than the other two estimates by multiplying by four.
PERT = (Pess. + (4xML) + Opt.) / 6
In doing this, we have provided a range of estimates, and perhaps the actual duration will reside within that envelope. We have a download excel sheet demonstrating how this works available.