In this paper, Joe introduces us to the application of Bayesian theory to assess how we are doing on our project after we have started — in fact at our first milestone. Essentially, Bayes consists of two pieces: A prior probability, and some new information. Combining those two gives you a new probability. The crucial observation is that this can be applied as an iterative process, with each new probability becoming the prior probability for the subsequent iteration. In essence, it is a bootstrapping process that can be repeated through the project's life span.
True, that Bayesian theory is subject to the criticism: "Where do you get the first (i.e., prior) probability in the first place?" If that is nothing better than a poor guess, then garbage in, garbage out is claimed. However, experience shows that in most cases the result after several iterations is insensitive to the original estimate anyway, because the new data quickly adjusts the probability for us if the subsequent tests are of good quality. But "good" project managers interested in applying Bayesian theory will have made sure that they have a good baseline estimate and plan to work with in the first place, so that should not be a problem.
Note that while this paper by Joe provides us with a recipe for updating, his paper published last October tells us how to be sure of the first estimate. That makes these two papers a complete Bayesian package, happily published in the right order!
All projects begin with unfettered optimism. We make plans, obtain funding, and recruit a team; occasionally the new project manager makes an estimate of the probability of success. We proposed a model and a methodology in Predicting Project Outcomes to do so.
Several months later the first major milestone date arrives, either with or without the promised deliverable. This milestone provides a go or no-go gate, and it's crucial to estimate the new probability of success based on performance to date. Experienced project managers (PMs) frequently make the right call, but not always. Less experienced managers often get it wrong. For that cohort, tools that rely more on analysis than on instinct are helpful.
There is a proven technique for updating a probability estimate with new data. It was invented in the 1750s by Thomas Bayes, and it has had a renaissance in the last 50 years. Many PMs have some knowledge of probability and statistics, but they may not have been exposed to the Bayesian approach. We demonstrate how it can be applied to continually update the estimate of the probability of success as more performance data accrete. A Bayesian framework allows us to make better estimates as time goes on.