Your estimate is wrong in a direction you can predict

The error has a direction

People are asked how long something will take, they answer, and the answer is too low. Not sometimes — characteristically, and in the same direction, which is what makes it interesting. A random error would average out over many tasks and it does not. This tendency is well known enough to have a name, often called the planning fallacy, and the useful part is that a bias with a consistent sign can be corrected mechanically.

The reason it happens is that when you imagine doing a task you imagine the work itself, not the interruptions, the review cycle, the dependency that arrives late, the thing that turns out to be more complicated than it looked, or the two days you will be pulled onto something else. You estimate the path where nothing goes wrong, which is a path you have almost never actually walked.

Look at the outside, not just the inside

The strongest correction is to stop estimating from the details and start estimating from history. How long did the last three things like this take? That number already includes interruptions and surprises, because it is a record of what actually happened rather than a projection of what should.

This feels like cheating, and people resist it because every task seems uniquely straightforward from the inside. It is worth noticing that the last one seemed straightforward too. Keeping a rough log of estimated versus actual for a few months is tedious and produces a personal correction factor that is more accurate than any general rule of thumb.

Decompose, but know what it does

Breaking work into pieces and estimating each does improve accuracy, mainly because it forces you to notice steps you had silently skipped — testing, review, deployment, the approval that takes four days. What it does not fix is the optimism inside each individual estimate, and it introduces a smaller opposite problem: a long list of small items each rounded down accumulates a large shortfall.

It also helps to estimate ranges rather than points. "Three to five days" is more honest than "four days" and it communicates something a single number cannot, which is how much you know. A wide range is not weakness; it is information that lets whoever is planning around you decide whether to reduce the uncertainty first.

What to do when it slips

Every estimate is eventually wrong, so the more valuable skill is what happens next. The rule that matters is that the value of a slip report decays fast. Reported early, it is a schedule adjustment somebody can absorb by moving something else. Reported at the deadline, it is a failure that other people now have to explain to their own stakeholders, and it is the second version that damages trust.

What people fear is looking incompetent, so they wait, hoping to recover. Occasionally they do, which is what keeps the habit alive. Mostly they compress a small problem into an unavoidable one. Say it as soon as the probability is real, with what you know and what you are doing: "this is likely to slip by three days, I will know for certain on Wednesday, here is what would help."

And distinguish between a slip and a scope change, because they are treated very differently. Work that grew because someone added requirements is not a missed estimate, and letting it be recorded as one is a small injustice that accumulates. Say plainly what was added and re-estimate rather than absorbing it silently.

Padding, and why it is not a strategy

The obvious response is to inflate everything, and it half works. The problem is that padding applied uniformly and privately gets discovered, at which point your numbers stop being believed and get discounted by whoever is planning — which pushes you to pad more. That is a stable and unpleasant equilibrium, and it is common.

The better approach is visible buffer rather than hidden buffer. Give a range, name the assumptions the estimate depends on, and say where the risk is. "Four days if the data access is already set up, closer to eight if it is not" is far more useful than a padded seven, and it invites the other person to remove the uncertainty rather than argue with the number.

It also, over time, makes you the person whose estimates are trusted, which is worth a surprising amount. Being reliable about time is rarer than being fast, and in most workplaces it is valued more once anyone notices.