Two Approaches to a New Estimate
In most cases the initial estimate of the probability of success is at least 50%, else the project wouldn't have been funded. You can do a crosscheck on your intuition with the Predicting Project Outcomes approach.
As the project progresses, we monitor performance against plan and constantly reevaluate the estimate. There are two approaches. The first starts with a clean sheet of paper each time, estimating as though it were a new project consisting of the remaining work. Based on what you learned, you would inject new numbers for team strength and process effectiveness in the PPO approach.
The second is to retain the last estimate and update it based on the new data obtained since that estimate was made. That estimate encapsulated all the knowledge about the project at the time, so it is silly to throw it away. The updated estimate consists of the weighted average of two components: The first is the baseline (or current) estimate, and the second is an adjustment based on the reliability of the new information.
After the process has gone through several iterations, the current estimate is the baseline plus multiple weighted historical data points, and the adjustment is based on the latest data point. At any time, the current estimate contains all we know about the project. This second approach is shown schematically in Figure 1.
Figure 1: Schematic of the second approach
We advocate using this second approach. For those who are unconvinced, consider this scenario: Suppose you are given a selection of 100 numbers each week and asked to compute their average. You are also asked to compute the "running average" from the beginning of time. To compute this running average at week 10, would you take the average of all 1,000 numbers, each of which you had to store?
Or would you take the running average after week 9, multiply it by 9, add the week 10 average to it, and then divide the total by 10? Both methods yield the same result. The second method involves less storage and far less work.
In this averaging example, the old running average has a weight of 9 and the new week10 average has a weight of 1. The new data is only 1/9 as important as the old data. Of course, weighting new project information is not so simple. Bayes' Theorem tells us how to do it.
