The #NoEstimates discussion now has soemthing to talk about. There is a presentation titled Re-Estimating the Value of Estimating. While there little here that is applicable in our software intensive world, I had a smile when I cam to this slide.
As one who works programs where Monte Carlo Simulation is mandated on a montly basis of cost and schedule Estimates to Comlete (ETC) and Estimate at Completion (EAC) and probabilistic estimates of the Technical Performance Measures compliance at some point in the future, this diagram is simply uninformed.
Noone working in the estimating business would ever draw the picture on the left. Such a symetric probability distributin is physically impossible on actual programs. Mike Cohen had a similar picture in his Agile Estimating book. An email corrected that misunderstanding. But it seems those in the #NoEstimates are still misinformed about the underlying stochatic processes of software development, hardware development, or general system engineering.
The activities of probabilistic forecasting when spending other peoples money, many times billions of dollars of other peoples money, if highly developed. Let's start with the GAO Cost Estimating and Assessment Guide. This is a general purpose guide for making estimates. NASA's Cost Estimating Handbook is another guide we use on our programs. The Air Force programs have their own Cost Risk and Uncertainty Analysis Handbook.
So Here's the BIG Questions (That Have Yet To Be Answered)
- What is the domain where #NoEstimates is applicable?
- What is the size of a project where you'd find a customer willing to have software developed without an Estimate At Completion or an Estimate To Complete?
- What is the Value at Risk while performing development without an ETC or EAC? This means what is the historical overage of cost or under delivery of features on the need date?
In the End
Starting with the picture above would be a non-starter when we spending 10's to 100's of millions of other peoples money. A moderate SAP installation is $20M. A typical health insurance ERP system is $150M. A suite of flight software for a manned spacecraft is $500M. You see the point.
You can likley to #NoEstimates when the value at risk is low. In that domain, the #NoEstimaters probably have a point. It's not that valuable to make estimates when the risk is low and the value is small.
Maybe that's the domain and context for this approach.