Before you start
Estimate Accuracy Analyzer
Find out how far off your team's estimates usually are, for which kinds of work, and what that history says about the next estimate.
Open a list of finished work with each item’s estimate and actual, in the same unit: hours from time tracking, days, or cost. A Jira export with Original Estimate and Time Spent works as it is. It is read inside your browser; nothing is uploaded.
Finished work
Drop a spreadsheet here
One row per finished item, with its estimate and its actual in the same unit (hours, days or money). A type of work column lets the page compare kinds of work; who estimated and when it finished are used if present.
Your own history is the best estimate of your next estimate
People estimating a job picture it going well; that is how estimating works, and it is why estimates run optimistic almost everywhere. The correction with the best record is to look at what happened last time: take the jobs already finished, see how their actuals compared with their estimates, and read the next estimate through that. Forecasters call it reference class forecasting. The reference class here is yours: this team, this kind of work.
What is measured
For each finished item, actual divided by estimate. The page reports the median, which one wild overrun cannot drag, the share that ran over, the share that landed within 10%, and the range the middle 60% fell in. By type of work, because design and backend work are rarely off by the same amount. Whether the largest estimates overrun by more than the smallest, which usually means big items need breaking down. And whether estimates are getting better or worse over time, if the file has finish dates.
Reading a new estimate
Type an estimate and a type of work, and the page multiplies it by the ratios your history reached at 50%, 80% and 90%. Those are the figures to plan and commit with: the first is a coin toss, the second is what most teams mean by a date they can keep.
What this does not do
It measures how estimates turned out, not why. Work whose scope grew after it was estimated counts as an overrun, so keep the change log in view. It does not rank estimators; differences between people are usually differences in the work they were given. A short history gives rough figures, and the page says when the history is short.