When your estimates are always wrong in the same direction

You said the reading would take an hour. It took three. This is not new information — the last one also took three, and so did the one before. What is strange is that the next estimate will also be an hour, and you will believe it while saying it.

The part that is counter-intuitive

The obvious explanation is inexperience, and it is wrong. The effect survives experience. In the classic work on this, people underestimated how long their own tasks would take even when they knew they had underestimated before, and even when they were reminded of it. Being told about the pattern does not correct the pattern.

The reason it holds is that estimating by imagining the task produces a particular kind of story. You picture the work going as intended: sitting down, the file opening, the argument being clear. You do not picture the hour spent finding the right version of the document, because there is nothing to picture — obstacles are specific and unforeseeable one at a time, while being entirely predictable in aggregate.

So you get a number that describes the task under good conditions. Good conditions are not the average condition. They are the best case, and the best case rarely happens twice in a row.

Estimate from records, not from imagination

The alternative is boring and it works. Instead of asking how long this will take, ask how long the last three like it took.

Doing this needs a record, which is the only real cost. For two weeks, write down what you started, what you finished, and the elapsed time. Three columns, anywhere. After a few entries you will have your own ratio between what you predict and what happens, and it tends to be reasonably stable per person and per type of task. Your own number beats a rule of thumb you read somewhere.

Estimate the whole chain, not the core task

The second half of the problem is scope. When you estimate an essay, you estimate the writing. The essay is also finding the sources, discovering two are behind a paywall, the reference list, the formatting the department insists on, the upload that fails once, and the twenty minutes of not starting at the beginning.

The core task is usually estimated fairly well. It is everything hanging off it that goes missing. A quick way to catch this: before you start, name the first action and the last action. If the first action is open the document and the last is press submit, count the steps between them. Most of the missing time lives there.

Where fast optimism pays and where it costs

It genuinely pays. Someone who says yes, I can have that by Friday gets asked, gets included, and starts things that a more accurate person would talk themselves out of. Underestimation is part of how anything difficult ever gets begun; if everyone costed their work correctly in advance, a fair amount of good work would never be attempted. It also protects you from the paralysis of seeing the true size of something all at once.

It costs when the estimates are chained. One optimistic guess is absorbed by an evening. Six of them stacked across a fortnight produce a week that cannot physically fit, and the failure does not arrive as one large problem — it arrives as everything being slightly late, which is harder to explain and harder to recover from. The bill is also paid by whoever is waiting on you, which is where the social cost sits.

What the setting can do

A great many deadlines are set by people who have never done the task, or who did it years ago under different conditions. That is not malice; it is missing information, and missing information can be supplied.

Negotiate on evidence rather than on feeling. This took me eleven hours last time and I have two of those due the same week is a different conversation from this feels like a lot. Ask how many hours the deadline assumes — often nobody has said it out loud, and asking makes the assumption visible and correctable. Then ask for a structural change rather than a personal favour: staggered deadlines across modules, a published expected workload per assignment, or a checkpoint partway through instead of a single end date. Course reps, module leads and managers can act on all of those, and they are much easier to grant than individual extensions.

Small things to try this week

1. Before your next three tasks, write your estimate down. Afterwards, write the actual. Do nothing else with it yet.
2. On the next thing you plan, list every step from first action to submission, then estimate the list rather than the task.
3. Ask one person who set a deadline what they think it takes. The gap is usually informative for both of you.

If your estimates stay wrong after weeks of recording, that is fine and still useful — you now have a number you can apply, which is what you needed anyway.