Agency rework: which correction should you investigate first?

To measure rework in an agency, start with corrections that repeat work to meet something already agreed. Record their minutes, group them by what went wrong and keep the evidence that could help investigate their origin. Three lengthy corrections can take more effort than four short ones. That distinction matters before adding another review round.

For an agency, the comparison starts with the work defined in the accepted proposal.

The log below lets you practise that decision. All fifteen episodes and their evidence are fictional. They do not describe clients, the productivity of Frihet’s team or a normal industry rate.

A defect needs a reference

An image measuring 1080 × 1080 pixels fails the assignment if the accepted requirement called for 1200 × 628. Comparing two valid compositions during a planned review is a different activity. So is rebuilding a prototype to learn a technique, or preparing an additional format that has just been requested.

The distinction prevents every iteration from being treated as waste. If you cannot identify which requirement was accepted before the work began, leave the classification unresolved. Do not turn a later preference into a defect attributed to the person who did the work.

ASQ’s cost of quality framework distinguishes prevention, appraisal, and internal and external failures. It includes rectifying errors among internal failures when they are detected before delivery. We use that distinction to select corrections here. The log does not calculate the total cost of quality, and it is not an ASQ method designed specifically for agencies.

A log of fifteen internal episodes

Each row records active minutes spent on that episode, excluding waiting time. The unit is the person-minute: two people working for ten minutes add up to twenty. Count the same effort only once, even if the problem is detected at several stages. Record routine inspection and cause-investigation time separately; do not add them to correction minutes. All episodes are detected or agreed before delivery. The log is a selection created for teaching: it does not contain all the work an agency performed during a period.

The ten corrections total 450 minutes. Planned reviews account for 105 minutes, exploration for 120 and the added format for 90. The log totals 765 minutes. The five activities excluded from the defect analysis total 315 minutes and remain work: a different label does not remove them from the project’s actual hours.

Dividing 450 by 765 would describe the composition of this selected log. It would not give the agency’s rework rate, because the other production hours are missing. We also have no revenue, hourly cost or margin data here. Turning the result into a money figure would invent information.

Three exports outweigh four version errors

To choose a problem to investigate by effort, we group only the ten correction episodes. The denominator for the percentages is 450 minutes. This ranked table provides the values for a time-based Pareto chart:

The first row is calculated as 270 / 450 × 100 = 60%. Cumulative shares use unrounded minutes; percentages are displayed to two decimal places. There is no requirement for the result to follow an 80/20 split.

If you count episodes, incorrect versions come first: four against three. If you add up minutes, incorrect dimensions lead: 270 against 75. Neither ranking explains the cause. They answer different questions. ASQ’s Pareto procedure allows frequency, time or another measure to be ranked from largest to smallest. Choosing the measure is part of the decision.

Recorded effort is the focus here, but it should not determine the priority of every problem on its own. An infrequent defect that prevents delivery deserves attention even if it takes few minutes to correct. The example does not measure severity or exposure; do not use it to dismiss either.

What you observe is not yet what caused it

In E01, we know the output has incorrect dimensions. “The wrong preset was selected” is a hypothesis. “Found during the pre-delivery check” identifies the detection stage, not where the problem was introduced.

To investigate that episode, you would keep the accepted reference, the configuration actually used, the output file and the sequence of versions. ASQ’s explanation of root cause analysis describes approaches that collect evidence and reconstruct events. Rewording a symptom as a cause does not do that work.

In the fictional examination of E01, reference F01 requires 1200 × 628 and configuration F02 records the correct preset. Output F03 is still 1080 × 1080. Trace F04 shows that the process used an old cached template. F05 represents a repeat using the current template and the same preset, producing the intended dimensions.

That evidence is part of the teaching scenario, not a software execution or an experiment conducted with an agency. Within the scenario, it refutes the explanation that someone manually selected the wrong preset and supports investigating the old template’s use. It does not establish that E02 and E03 share the same origin. A common symptom is insufficient: their configurations and traces still need examination.

Prepare the next check

Alongside the ID and minutes, a useful record keeps the accepted requirement, the observed problem, the detection stage, the hypothesis and the missing evidence. That lets you return to a specific episode. Avoid labels such as “carelessness” or “poor communication” when they do not describe something you can examine.

For this log, the defensible decision is to inspect the evidence for E02 and E03 before recommending preset training or changing the whole process. It is not a promise to recover all 270 minutes. The log cannot tell you how much effort would be prevented, how much fixing the process would require or whether other failures would appear.

Start with a period and a type of delivery you can observe consistently. Keep the cases you cannot yet classify, too. The first useful result is modest and specific: a priority problem, a hypothesis labelled as such, and the next evidence needed to accept or reject it.