Estimation Range
Expressing an estimate as an interval with a stated confidence rather than as a single number, because a point estimate misrepresents what is known.
A single-number estimate is read as a commitment regardless of caveats, and it conceals the one thing the recipient most needs: how uncertain it is.
The practice: give a range — "six to ten weeks, and I'd be surprised if it exceeded fourteen" — and say what drives the spread. That is more useful for planning than a point, and it is more honest.
What makes estimates better, in order of effect:
Decompose. Errors on small pieces partially cancel; errors on one large guess do not.
Use historical data from comparable work rather than imagination. Reference-class forecasting outperforms expert judgement consistently, and the data usually exists.
Estimate the unknowns separately. Work that requires discovery should be time-boxed as a spike, not estimated as if it were understood.
Name what would make it longer — the dependency, the unfamiliar component, the approval — which converts an estimate into a risk conversation.
The biases to correct for: the planning fallacy, where people systematically underestimate even knowing they do; anchoring on a number someone mentioned first; and optimism about the happy path, which omits the integration, the review, the environment and the rework.
And the professional point: estimates are not commitments. Where a commitment is required, it should be built from a range with contingency, and the difference should be said out loud.