Start with the complete task

A draft is useful only in relation to a task somebody needs to complete. For this illustration, the task is to prepare an action list from fictional meeting notes and have a reviewer verify each commitment. The output must preserve owners, dates that were actually agreed and unresolved questions.

The page compares two synthetic assisted paths with the same fictional thirty-minute manual baseline. It does not estimate typical performance. Its purpose is to show how preparation and review can change the arithmetic even when drafting remains fast.

Define the comparison

Use the same source material and acceptance conditions for both approaches. The manual baseline must include the review required for a usable result, and the assisted path must include its preparation and correction. Otherwise the totals describe different tasks.

Keep setup work separate from recurring work and note which steps the example excludes. If a real trial later requires training, source cleanup or approval, record that work rather than treating it as free. The fictional totals on this page cannot settle an actual investment decision.

See where the minutes go

In the clear-source example, preparation takes six minutes, drafting three, review twelve and correction four. The assisted total is twenty-five minutes against the thirty-minute manual baseline. The five-minute difference is arithmetic for this example, not measured time saved by a tool.

In the conflicting-source example, preparation takes eight minutes, drafting three, review twenty-four and correction thirteen. The assisted total is forty-eight minutes, which is eighteen more than the same baseline. Switching examples changes the displayed stages and total; it does not change the underlying quality requirement.

Ask why review is difficult

Imagine that the fictional notes contain two owners for the same action. A reviewer may need to reconstruct the sequence of messages before deciding which ownership statement is current. That effort is part of the task even if the draft itself appeared immediately.

Record the specific cause of the review work. Missing source versions, ambiguous status labels and unsupported commitments call for different corrections. An explanation such as “the answer was not good” is too vague to guide another trial.

Keep quality separate from speed

A correct total does not establish a correct answer. Record omitted actions, invented dates and permission failures separately from the timing comparison. A short task that produces an unauthorised external action should not be celebrated because its elapsed time is small.

Use a case that the workflow should decline to answer, such as a hotel recommendation absent from the fictional policy set. A clear statement that the sources do not establish a choice is more useful than a fast invented recommendation.

Make the handoff explicit

Identify what the next reviewer receives: the proposed output, its source references and the questions still unresolved. Do not leave that person to infer whether an action is proposed, approved or already completed. The labels should make those states distinguishable.

For the fictional meeting case, an unassigned action remains unassigned. A summary should not allocate it to the nearest named person. The receiving reviewer needs to see the missing ownership decision so an authorised person can resolve it.

Revise one part of the exercise

Keep the source case constant while changing one part of the workflow when practical. For example, label the current note more clearly, then repeat the review. Record whether the correction addressed the previously observed failure rather than judging only the new answer in isolation.

If several parts change together, say so. You may still learn from the result, but the cause of the difference becomes less certain. Preserve the earlier output, source set and acceptance conditions so another reviewer can understand the comparison.

Decide what the evidence supports

End with a bounded recommendation: stop, revise or try another approved case. Name the evidence supporting that choice and the uncertainty that remains. A small fictional exercise can reveal obvious problems but cannot establish performance across a whole organisation.

Before a workplace trial, involve the owners of the sources, review process and proposed action. Agree on permissions and the stop condition. This field note organises questions for that discussion; it does not certify a workflow or substitute for the responsible people’s decision.

Notes on the illustration

The two timing examples were authored for this learning page. They are not sampled observations and have no population, confidence interval or industry comparison. The bar lengths represent whole minutes for the four named stages, and the totals are their sums.

The offline edition contains the complete explanatory text and both sets of example values. It has no tracking, forms or external assets. To explore different values interactively, open the linked work review workbook and compare the complete task rather than the draft alone.