Blah.Blah.Blog

This estimate came from a multi-day workshop to thrash through some the issues confronting us for the second major phase of a development effort we had just moved into production. We had been locked up in a room with the PM and a couple of BAs ploughing through a 20 page list of one-line requirements that would be fleshed out/decomposed into the story cards for this next phase.

For each line item, we attempted to gather as much extra information from the BA, equate the litem with work from the first phase where possible (using our list of actual time spent against each story), and then generate a high level estimate in order to give an overall figure to the development effort for the entire phase.

Along with the estimate came a level of precision, which varied wildly against our ability to equate the line item with previous work, and it's associated actuals. And herein lay our problem...

We had been using a percentage figure to give shape to our precision values. In other words, a line item which was just another instance of something we'd already developed would have an estimate along the lines of 2 days +/- 10%. For the completely foreign items, it may be something like 2 weeks +/- 100%.

And it never occurred to me until a collegue who has probably forgotten more about project management and estimation than I will ever know pointed out the implicit assumption whenever you use percentages as units of precision. A percentage is inherently bounded. By definition, a percentage cannot exceed 100 in either the positive or negative direction. Sure, people do abuse the concept with terms like 110%, but the fact remains that people naturally think of a percentage not being able to exceed 100 - as if the mere application of the term has suddenly implied a mathematical constraint to the reality where none existed previously.

The big problem waiting to happen with our use of percentages is the reaction from all and sundry when one of the unknown items turns out to be much, much bigger than we had anticipated and a 2 week +/- 100% estimate becomes 6 person weeks of implementation. Suddenly our worst case estimated scenario (4 weeks) was still way too optimistic.

Thankfully, we never had to stand by these estimates, so our mistake never had a chance to bite us further down the path. Hopefully, everyone involved has learnt from the potential calamity our actions could have caused. Thanks to the excellent "Walzing With Bears", I am now much better equipped and prepared to deal with these situations.

Note: we had not made any attempt to document our level of estimate accuracy in this process. As detailed in Roberts' log, adopting this approach could also have helped alleviate any false expectations set though the use of percentages for our precision units.