3 ms·
> Say you want to know how long it will take to develop some project (people famously always get this wrong due to optimism bias). Actually, I find that people
by bsder 9y ago
> Say you want to know how long it will take to develop some project (people famously always get this wrong due to optimism bias).
Actually, I find that people are, in fact, quite reasonable at predicting projects. If I ask for two points, 50% and 85% probability of completion of task, and then I put everybody's tasks together I get a really remarkably good prediction of the project timeline.
Anybody doing a task for enough time gets a really good feel for schedules. Oddly. Sometimes you even wind up with great heuristics ("2 weeks per engineer per schematic" is one of my favorites).
The problem is that management wants a 95% probability point. Unfortunately, the variance/standard deviation goes flying to the moon, and it is IMPOSSIBLE to convince people that is correct in spite of mountains of evidence both theoretical (really, every statistics book ever) and practical (gee, all of those "late" projects could be the fact that your people are incompetent/liars OR it could be that you are willfully abusing statistics. Things that make you go, "Hmmmmmmm.")
So, that 95% probability point is WAY longer than anybody is willing to agree to. From a political perspective, it's better to play schedule chicken and pray that a lower probability point comes home, that you can blame somebody, or that everybody is now sufficiently invested that they are willing to continue to the real 95% point.
I actually did this for 2 different large projects where I put the estimates together, ran some random Monte-Carlo modeling of different task chainings, and posted the timeline on my wall. Everybody attacked the schedules--engineering and management. The projects came in at the 62% point and the 75% point. And that was with changes (which always cause schedule extension). Both of these are less than a SINGLE standard deviation late--by quite a bit (Yes, I know I'm horribly abusing the notion of "normal distribution", but I have no hope of explaining anything even slightly more complicated to business types).
Management asked me not to do this for the third project because people were actually paying attention. I left shortly thereafter.