Retrospective Estimation
Reconstructing the day from memory at the end of it. The cheapest method, systematically wrong in known directions, and useful once you know which.
At the end of the day, you write down where you think it went. It takes three minutes and it is wrong in predictable ways. When this method moves from a personal experiment to a team process, read the agency example can reduce manual reconstruction, provided everyone knows what is recorded and the organisation keeps only the detail needed to answer the stated question.
The known biases
Long stretches get overstated. Three hours of work with interruptions is remembered as three hours of work.
Fragments vanish. The eleven minutes looking for a file, the four small interruptions, the fifteen minutes between finishing and starting.
Unusual days are remembered as typical, and typical days are not remembered at all.
Unpleasant tasks feel longer than they were; absorbing ones feel shorter.
And the total is optimistic: people consistently account for fewer hours than the day contained.
Why it is still worth doing
It costs three minutes and produces something.
It is sustainable indefinitely, which the diary is not.
And the direction of the errors is known, which means a corrected estimate is better than no data at all.
The correction is crude: expect fragmented time to be larger than you recorded and deep work smaller.
The most useful version
Estimate first, then check.
Write down where you think last week went, before looking at anything. Then run one week of sampling or capture.
The gap between the two is the finding, and it is more informative than either figure alone.
It is also the exercise with the best evidence behind it in this whole collection, because the estimation error is the thing most reliably demonstrated.
Doing it at the right moment
At the end of the working day, not the next morning.
Overnight loses the fragments entirely and compresses the day into its two or three main activities.
Five minutes at the point of stopping, in whatever form is quickest.
What it is adequate for
Noticing a trend over months: the estimate is wrong consistently, so changes in it still mean something.
Spotting a week that was unlike the others.
Keeping some record where any other method would be abandoned.
And answering "roughly how much of my week goes to this" within an hour or two, which is enough for most personal decisions.
What it is not adequate for
Billing. A reconstruction is a different kind of record from a contemporaneous one, and where the distinction matters it matters a great deal.
Estimating future work. Estimates built on estimates compound the same optimism.
And finding fragmented time, which is exactly what it cannot see.
Improving it slightly
Use the calendar as a prompt rather than working from memory alone: it anchors the fixed points and the reconstruction fills between them.
Write the interruptions you can recall separately, which partially corrects the main bias.
And note the day's rating — good, ordinary, bad — because over months that is frequently more useful than the hours.
What to check
Are you writing it at the end of the day or the next morning?
Do your estimated totals account for the whole working day, or do two hours go missing?
Have you ever checked an estimate against a record?
And are you using an estimate for something that needs a record?
Estimate first, then check
The most useful version is not the estimate itself but the comparison: write where you think the week went, then measure one week.
The estimate is contaminated once you have seen the data, so the order matters.
Worth keeping in mind
Write it at the end of the working day rather than the next morning. Overnight loses the fragments entirely and compresses the day into its two or three main activities, which removes most of what was worth recording.
What it cannot support
Billing, because a reconstruction is a different kind of record from a contemporaneous one.
Also estimating future work, since estimates built on estimates compound the same optimism. And finding fragmented time, which is precisely what memory cannot see.
In short
Use the calendar as a prompt rather than working from memory alone: it anchors the fixed points and the reconstruction fills between them.
Noting the interruptions you can recall separately partially corrects the main bias. For a complementary perspective on work, measurement or planning, consult the American Psychological Association.