Manual Entry

Someone typing what they did is the default method in most organisations. Its weaknesses are well understood and several of them are fixable.

What it is good at

Attribution. A person knows which matter the work belonged to; no automatic system infers this reliably.

Judgement. Which of three overlapping tasks the thinking time belongs to is a call only the person can make.

Narrative. Where a client expects a description, it has to be written by someone who knows what happened.

Low cost. No agents, no hardware, no capture infrastructure.

Nothing intrusive. The record contains what the person chose to put in it.

What it is bad at

Duration accuracy, especially reconstructed.

Fragmented work, which cannot be reconstructed at all.

Completeness. Small tasks and internal work are systematically under-recorded.

Consistency between people, who draw boundaries differently.

Timeliness, which is the root of most of the above.

The delay problem

Accuracy falls sharply with delay, and delay is the variable you can actually change.

Same day is much better than same week.

In the moment is better still, though it interrupts.

A reasonable target: most entries made within one working day. Measure the proportion; it is the single most useful data quality figure in this field.

Reduce delay by reducing friction, not by chasing people.

Reducing friction

Few categories. A person should be able to hold the list in their head. Hundreds of project codes produce guessing.

Recent and frequent items first, since most entries repeat.

Entry from where the work happens — the editor, the ticket, the phone — rather than a separate system.

Pre-fill from the calendar for meeting-shaped work, with the person confirming rather than typing.

No mandatory fields that do not matter. Every required field is a reason to postpone.

A one-minute daily entry beats a twenty-minute weekly one, on accuracy and on how people feel about it.

Approval

Manager approval is common and is mostly theatre unless the manager has information the system does not.

A manager who approves forty timesheets on a Friday is not checking anything.

Where approval exists, make it exception-based: flag the unusual — totals far from expected, large single entries, retrospective edits — and let the rest pass.

Measure the rejection rate. Zero means it is not a control.

Where it should not be the only method

Where the legal duty requires an objective record of daily working time, a self-reported weekly reconstruction sits awkwardly against the standard.

In that case the surrounding controls carry the weight: prompt entry, worker confirmation, protection against employer alteration, and an audit trail on changes.

Or a clock-in method covers the statutory record while manual entry covers project attribution, which is the arrangement most organisations end up with and is worth designing deliberately rather than drifting into.

Exception-based approval

Approval that means something, rather than forty clicks on a Friday.

Flag the unusual: a total far from the person's range, a large single entry, a retrospective edit, a missing day.

Let everything else pass.

Measure the rejection rate. Zero means the step is not a control.

Measure time per approval. Seconds means nobody is looking.

A manager attending to six exceptions attends to them; one facing forty timesheets attends to none, and the second arrangement provides assurance that is not there.

Reducing the category list

The intervention with the highest return and the least cost.

Filter to the person's own work: their projects, their recent entries, their team.

Under twenty active options visible at any time, whatever the organisation's full catalogue contains.

Recent and frequent first, since most entries repeat.

Close finished projects, or the list only grows.

A person who has to look something up will postpone, and postponement is the mechanism by which accuracy is lost.

Turn the principle into a test

For an example that can make this requirement testable, consult review the agency workflow. Treat the page as a starting point rather than proof: reproduce the workflow with real roles, exceptions and permissions before relying on the result.