Estimation Fails Because Nobody Separates the Two Questions
Estimation disputes are usually not about effort. They are about a hidden substitution: someone asks how long a task takes, receives an answer about focused working hours, and interprets it as a delivery date. Those two numbers differ by a factor that depends on meetings, interruptions, review latency and unrelated production incidents — none of which the engineer was estimating.
Ask both questions explicitly
How much work is in this, if nothing interferes? And given everything else in flight, when will it realistically be done? Answered separately, both are answerable. Combined, neither is. Discussions of forecasting practice, such as those published at the planning coverage on this site, keep returning to this separation as the cheapest available improvement.
Use ranges, and say what they mean
- A single number implies precision the estimate does not have
- A range communicates uncertainty without requiring a statistics lesson
- State the assumptions the estimate depends on, in writing
- Name what would make it wrong — unclear requirements, an untested dependency
Historical data beats intuition
Teams that record how long similar work actually took develop forecasts far better than any estimation ritual produces. Counting completed items per week and projecting forward is unglamorous and consistently more accurate than summing optimistic guesses, because it includes the interruptions automatically. It also removes the incentive to argue about point values, which is where most of the meeting time goes.
The organisational half of the problem
Estimates degrade when being wrong has consequences. If an underestimate produces pressure and an overestimate produces suspicion, engineers learn to pad quietly and everyone stops trusting the numbers. Treating estimates as forecasts to be updated — rather than commitments to be enforced — is what keeps them informative. A forecast that changes as information arrives is working correctly, and a plan built on the assumption that it will not is the actual source of the missed date.
