In a recent email exchange, the paper by Todd Little showing projects that exceeded their estimates was used as an example for how porrly we estimate, and ultimately one of the reasons to adopt the #NoEstimates paradigm of making decisions in the absence of estimates of cost, schedule, and the probability that the needed capabilities will show up on time and be what the customer wanted.
Sherlock here had it right. This picture by the way is borrowed from Mike Cohn's eBook of 101 quotes for agile.
I've written about Little's paper before, but it's worth repeating.
It's very sporty to use examples of bad mathematics, bad management, bad processes, bad practices as the basis for something new. This is essential the basis the book Hard Facts, dangerous Half-Truths: Profiting from Evidence-Based Management. When we start talking about something new and disruptive in the absence of the data, facts, root causes, and underlying governance and principles, we're treading on very thin ground in terms of credibility.
Here's the core principle of all software development
Customers exchange money for value. The definition of value needs to be in units of measure that allows them to make decisions about the future value of that value. That value is exchanged for a cost. A future cost as well.
Both this future cost and future value are probabilistic in nature, due to the uncertainties in the work processes, technologies, markets, productivity and all the ...ilities associated with project work. In the presence of uncertainty, nothing is for certain - a tautology. There are two type of uncertainty - reducible and irreducible. Reducibe we can do something about. We can spend money to reduce the risk associated with the uncertainty. Irreducible, we can't. We can only have margin, management reserve, or a Plan B.
To make decisions in the presence of these uncertainties - reducible and irreducible - we need to estimate the uncertainty, the cost of handling the uncertainty, and the value produced by the work driven by these uncertainties. When we fail to make these estimates, the uncertainties don't go away. When we slice the work into small chunks, we might also slice the uncertainties into small chunks - this is the basis of agile and the paradigm of Little Bets. But the uncertainties are still there, unless we've explicitly bought them down or installed margin and reserve. They didn't go away. And what you don't know - or choose to explicitly ignore - can hurt you.
So Sherlock is right don't put forth a theory without the data.