Lynda Bourne has a nice post titled Prediction is Difficult. In the end she speaks about the human failings of predicting and the difficulties of getting senior management is accept these predictions.
First I'd like to suggest a change from prediction to forecasting. The forecast includes a range of possible outcomes and the confidence level on that range. This is one root cause of senior managements lack of understanding about business numbers. They assume (in my experience) that project numbers are like accounting numbers. They are not of course, but as project managers we don't make that clear by using ranges and confidence intervals on those ranges, so what else they to think.
How To Speak About Project Numbers
When we speak about cost, schedule, and technical performance it MUST be in the following manner:
- We have a 70% confidence in completing on or before September 13th, 2013
- Our cost for this project has a 70% confidence of be $1.6M or less
- The sustained throughput of the data channel will be 950Mb other the Time Triggered Gigibit bus with a 80% confidence.
How Do We Develop These Confidence Discussions?
The very first we DON'T, as Lynda suggests is ask some subject matter expert. SME is the least confident answer. The highest confidence answer in our domain is called Past Performance.
We've done the 17 times before and have a statistically sound basis of our estimate, and have a 80% confidence that this time the duration will be 18 weeks or less.
The source of this data may not be directly available, so the next best source is Reference Class Forecasting. A good place to look to see how this is applied in project management is Bent Flyvbjerg's "From Noble Prize to Project Management." (Follow the links from his wikipedia page to find many other discussions about Reference Class Forecasting).
All Point Estimates Are Wrong, Without a Variance and a Confidence on that Variance
Making single point estimates - this project will complete on August 17th of this year, is simply wrong. There is not one ounce of credibility in that statement. Without stating upfront what the range of possible finish dates, and confidence levels for each of those range (binning this is called) or have a Probability Distribution Function (PDF) that shows the confidence of an on or before completion, the production or forecast is simply bogus.
