The definition, expanded
When we estimate, we imagine the task going as planned: open the file, write, done. We don’t picture the missing password, the unclear brief, the second round of edits. Oddly, the bias is strongest for our own work. We are often better at estimating other people’s tasks, because we look at their track record instead of their intentions.
Where it comes from
Daniel Kahneman and Amos Tversky introduced the term in 1979, in work on intuitive prediction. Buehler, Griffin and Ross (1994) then tested it: students predicted when they’d finish their theses and other assignments, and most finished later than their own predictions, including their “worst case” ones. Kahneman’s proposed fix is the outside view, also called reference-class forecasting: base the estimate on how long similar tasks actually took.
How to counter it
- Before estimating, find two or three similar past tasks.
- Use their actual durations, not how long they “should have” taken.
- Break big tasks into pieces; small estimates are less wrong.
- Leave unscheduled slack in the day, at least an hour.
- After the task, compare estimate and actual, and adjust next time.
An example
Alex estimates “update the Shop product pages” at 60 minutes. Looking back, the last two updates took 95 and 110 minutes, because of image resizing and a pricing check. Alex sets the new estimate to 100 minutes and splits it:
- Resize images, 30 min #Shop
- Update copy and prices, 45 min
- Check on mobile, 25 min
The afternoon still has a free hour after 16:00 for whatever goes wrong.
Where the idea has limits
The fallacy isn’t universal. People sometimes overestimate very short tasks, and the effect varies with how familiar the task is. Padding every estimate by a fixed percentage can also backfire through Parkinson’s law: the extra time gets filled. Past data beats a blanket multiplier.
Handling it in Needt
Every task in Needt has an estimate in minutes, so the gap shows up quickly. When a task runs over, the rest of the day rebuilds around it and Needt shows what moved, so an overrun doesn’t silently push your evening. Once a week, look at the tasks that moved most in your weekly review and raise their estimates. Subtasks help too: three 30-minute subtasks are easier to estimate than one 90-minute blob.
Agent prompt: “Which tasks this week took longer than planned? Suggest better estimates for similar ones.”
