Note from the editor: We received the following feedback from member Charles Carroll.
I have a couple of suggestions:
1. Inform all hands (up as well as down) that there are three levels of estimation - ball park, reasonable and accurate and that these are based on the forecasting horizon. Ball park for the whole project or 6 months, reasonable for 1-3 months or the next phase/iteration and accurate for 2-4 week. Acceptable variances are established for each horizon.
2. Use a sliding baseline with earned value analysis based on the "accurate" estimation period. That allows people to learn as they go and you can suggest that estimation of short term work is an indicator of true expertise.
3. Don't confuse level of effort with duration. Estimate in terms of hours of work to be performed. Then introduce availability/effectiveness factors such as time off, attendance at "town meetings," participation of user subject matter experts, etc.
4. Introduce incentives (which means money) for accurate (not necessarily positive) estimation. This is measured over a significant period, say the most accurate estimators for the year. This brings with it the possible risk of padding to ensure "accuracy" so you will also need to be frank about competition with other teams. IT isn't used to this but people in other career fields are. If your estimations are accurate because they are twice as long as every other team doing the same thing, you get your bonus for estimating and your penalty for being the least competent team/person. Most IT people are salaried so the following might not work, but is a thought. An automobile dealer I know used to use standard (from the book) estimates for LOE on repairs. If the mechanic completed the repairs earlier, he got paid for the book estimate. If the car came back, an ombudsman determined whether or not it was for shoddy work. If it was, the mechanic had to complete the repair at no cost to the dealership. This dealership attracted the best mechanics in the area.
5. Paragraph 4 really speaks to using a feedback system with continual posting of results. It has to be accompanied with root cause analysis when an estimate is significantly off. This doesn't have to be a never-ending analysis. Analysis is followed by mentoring and referral to the most accurate estimating team (again, these folks get some type of incentive to help their fellow teams improve).
6. Among the reasons for inaccurate estimation is fear. If I estimate accurately and an unlikely or rare event keeps me from meeting my schedule, I will get hammered. In this case the project or team lead has to support his or her team. Critical events such as non-receipt of equipment, no-shows by non-team participants, power outages, etc. need to be tracked and reported on a frequent (weekly) basis. These events should be subtracted from the estimation. Yes, Mr VP, your expert couldn't make the worksession because she had an unscheduled, important meeting with a client. We understand that. We just want you to know that this will slip task X three days. No hostility on our part; just the facts. The presence of a strong social contract between management and technicians/estimators will remove much of the fear incentive to pad estimates.