What Estimates Mean

When someone says a task will take two weeks, what number are they giving you?

Most people hear “two weeks” and think: that’s how long it will take. But that’s not what the estimate means. What they’ve given you is a median — the point where half the outcomes fall on each side. It’s the most likely duration, not the expected one.

The difference matters because task durations don’t follow normal distributions. They follow lognormal ones. The shape is asymmetric: tasks can finish a bit early, but they can run very late. You can’t finish a two-week task in negative three days, but you can absolutely finish it in six weeks.

This asymmetry means the average (the mean) sits higher than the middle (the median). For lognormal distributions, the mean is roughly 1.6 times the median. Your two-week estimate? The expected duration is closer to three weeks.

This isn’t a planning failure. It’s a structural property of complex work.

When you’re estimating, you’re probably thinking about the most likely path — the version where things go roughly as expected. That’s the median. But averages include the long tail: the vendor who ghosts you, the integration that uncovers a deeper bug, the requirement that changes mid-flight. Those aren’t anomalies. They’re part of the distribution.

The pattern shows up most clearly in technical work — software, engineering, R&D — where unknowns compound. But it applies anywhere outcomes have long tails. Construction projects. Product launches. Regulatory approvals. Anything where “what could go wrong” has a longer list than “what could go right.”

Different people need different numbers from the same estimate.

The engineer giving the estimate is thinking about the median — the realistic target if things go normally. Their manager needs the mean — what to expect on average across a portfolio of work. The executive making commitments needs a higher percentile — the 80th or 90th — because missing a deadline has real consequences.

None of these people are wrong. They’re just asking different questions of the same distribution. The engineer isn’t sandbagging when they say two weeks and it takes three. The manager isn’t padding when they tell the board four. They’re reading different points on the same curve.

The practical insight: a single number hides a distribution.

When you give an estimate, you’re collapsing a range of possible outcomes into one figure. When you receive an estimate, you’re hearing the median but probably planning against the mean — or worse, treating it as a ceiling.

The fix isn’t better estimation. It’s better communication about what the number represents. “Two weeks most likely, but could stretch to four if we hit integration issues” contains more information than “two weeks” alone. Three numbers — optimistic, likely, pessimistic — tell you more than one ever could.

The gap between what you meant and what the number means is where schedules slip.

Connects to Library: Lognormal Distribution