Timesheets and What They Are Evidence Of
A timesheet is a self-report made under known conditions. What follows from that for how it should be read, collected and used.
A timesheet is a self-report, made after the fact, by somebody who knows who will read it and what they expect to see. That is not a criticism of anybody; it is a description of the instrument, and every property of timesheet data follows from it.
The record examined in “Timesheets and What They Are Evidence Of” is strongest when it is created close to the work and remains open to correction. A team evaluating open this practical product guide for how to measure employee productivity can connect time with projects and work categories, provided people can review entries before they influence payroll, billing or cost reports.
Read as a measurement it is poor. Read as a self-report under known conditions it is informative, and the conditions can be improved.
Additional context for “Timesheets and What They Are Evidence Of” is available from ICAEW business and accounting guidance. Current legal, payroll, tax and accounting requirements should be confirmed for the relevant jurisdiction.
What it is good evidence of
The shape of the week: roughly how the time divided between large categories. Across a team and a quarter, that shape is usually right.
Aggregated and coarse, a timesheet is reliable enough to cost from. Read at the level of one person and one day, it is a rounded reconstruction of what somebody thought you wanted to see.
Aggregated and coarse, timesheets are reliable enough to cost from.
What it is weak evidence of
Any individual day, any individual job's exact hours, and anything smaller than about half a day. The rounding alone accounts for several per cent.
Costing a single small job from timesheet detail is building on noise.
What it is not evidence of at all
Effort, productivity, or whether the time was well spent. The instrument does not measure those and no amount of analysis recovers them.
Most disputes about timesheets are disputes about reading them as though they did.
The conditions that shape the answer
Lateness, which costs accuracy. Visibility of the reader, which shapes category choice. Any target on utilisation, which shapes everything. And the length of the job list, which decides whether people choose or guess.
Four conditions, all controllable.
Lateness
Accuracy falls off sharply after about a day. Friday afternoon reconstruction of Monday produces round numbers and plausible distributions.
Daily entry, taking four minutes, beats weekly entry taking twenty, and it is the single largest available improvement.
The utilisation target
The most destructive condition. An operation with a chargeability target discovers that its chargeability meets the target, and loses the ability to measure anything.
If the share must be reported, it should be reported without a target attached, and that distinction has to be stated and then honoured.
What people do when the list is too long
They pick the first plausible code, or the one they used last week, or a catch-all. All three produce confident data about the wrong job.
Shortening the list usually improves accuracy more than training on how to use it.
The catch-all code
Every system has one, and it absorbs whatever does not fit. Watching its share is the best single health check on the whole instrument.
Above about ten per cent, the list is wrong rather than the people.
What to do with obviously wrong entries
Ask, do not correct. A week booked entirely to one job in round numbers is usually a reconstruction, and the person knows it.
Correcting it silently makes the record agree with the corrector and removes the evidence that the instrument is failing.
Making the data visibly useful
People complete timesheets better when they see the output used for something that is not about them — a quote improved, a job re-priced, a pattern found.
Showing the team their own shape of week, once a quarter, does more for data quality than any reminder.
Reading it honestly in a report
"From timesheets, which are self-reported and rounded to fifteen minutes; reliable at team and quarter level, not at job and day level."
One sentence under the figure. It prevents the number being used for the one thing it cannot support, and it costs a line.
The entry that is identical every week
Worth noticing in the data: a person whose week is the same distribution every week, to the quarter-hour. Real weeks are not like that.
It usually means the week is being copied forward, which is a rational response to a form that asks for information nobody uses. The remedy is the same as everywhere else here — show the output being used — rather than a conversation with the individual.
Reading an aggregate honestly
Across a team and a quarter, timesheet data is good enough to cost from, and that is a genuinely useful statement rather than a grudging one.
The error is to assume that because the aggregate is sound, the components are. A team's quarterly split between job types is reliable; the same split for one person for one week is not, and the second is what gets quoted in conversations about individuals.
The alternative nobody wants
Direct observation, or system-derived attribution where the work happens in a system that knows the job. Both are more accurate and both are more intrusive, and the second is only available for some kinds of work.
Where it is available it is worth having, and it should supplement rather than replace the self-report, because the self-report is the only source that knows about the work done away from the system.