As per one of the experienced project manager even a good project estimate using one of the widely used methodologies there is only 50% chance of it being correct. What it says is that if your estimates are even 50% correct on delivery, you should be happy and actually did a good job! This despite that Monte Carlo estimate is supposed to give you over 95% correct estimate and if you are using agile you don’t really need to make a full estimate because you iterate the development for multiple deliverables concurrently.
Sure there are tons of article on the ways to employ better delivery estimation debate with so very divergent views from each practitioner. The truth is that you deal with so many dynamic factors that are play at the same time; it is really an art to repeat and beat your estimates each time you take a plunge. So even if you have the past historical data on your side for a similar project type, a team as experienced (or inexperienced)
As the one which completed the project in the past, and you are using the identical technology in your project with same project methodology, you would face your own set of project issues which could take your project completely off track. In short, you can only plan and have mitigation approach for what you know and that should be done however as a project manager one deals with many different dynamics it is almost impossible to have all your risks managed and assumptions validated to be sure of more than 50% of project success.
The another prominent factor quite evidently came out through a recent survey is that when estimating a project, most managers tend to have narrow estimate ranges than what is required to have over 90% confidence. When asked to put estimate ranges for specific tasks, PMs came out with estimate ranges which were narrower than something required to have 90% confidence. In fact the estimate ranges were only comfortable for 40-50% confidence.
So when you ask someone for a range that provides 90% confidence, expect 30% confidence on average Reason? We are naturally hesitant to provide wide ranges-because we feel that narrow estimates are a sign of a better estimate. But narrow estimates are really self defeating unless you have specific data to support the narrow estimates.