The regional calculations were correct
One region had ninety complete order-review records out of one hundred. Another had 450 out of 900. Their rates were 90% and 50%. In 2019 I approved a combined report showing 70%, the average of those two percentages. Under the same completeness criterion, the pooled population contained 540 complete records out of 1,000. Its rate was 54%. Repairing the report cost ₹19 lakh.
The two regional calculations were not the error. Each described its own reviewed records correctly. The problem appeared when I allowed an average of the regional percentages to answer a question about the whole record population. The smaller region received the same influence as the larger one. Ninety records in one place and 450 in another became two percentage figures with equal weight in the summary.
That transformation was easy to overlook because the report still looked familiar. It contained rates, a combined line and a number between the two regional values. Nothing about the visual presentation announced that the unit had changed. A reader could reasonably interpret the combined line as the proportion of complete records across the reviewed population, which was the purpose the report was intended to serve.
I had approved that interpretation. Describing the incident only as a formula problem would leave out the judgement involved in accepting the formula's meaning. A data team can calculate an expression correctly while a leader permits it to represent the wrong thing. The accountability I retain is for the connection between the arithmetic and the question management was being invited to answer.
The criterion in this case concerned commercial-record completeness. It was not a product safety result or a technical acceptance decision. That context matters because a number can acquire a broader significance as it travels. Even after the aggregation was corrected, the report still needed to be understood as evidence about the defined records and criterion, rather than a general score for the quality of the entire enterprise.
The intended unit determines the combined question
An unweighted average of regional rates is not inherently meaningless. It can describe a question in which each region is deliberately given equal influence. That was not the combined-record question here. The first task was therefore to make the intended unit explicit, rather than declare one arithmetic operation universally right and another universally wrong. We needed to know whose experience the executive summary was supposed to represent.
If the question concerns all reviewed records under a common criterion, the underlying counts are essential. Adding complete records gives 540; adding the reviewed population gives 1,000. The resulting 54% retains the larger region's larger contribution to the record population. It does not make that region more important in every managerial sense. It simply preserves the unit specified by this particular measurement question.
That distinction helps avoid a second mistake: treating statistical weight as a judgement about organisational worth. A small region may deserve substantial management attention for reasons unrelated to its count in a pooled measure. It may face different conditions or have strategic significance. Those reasons should be discussed directly. They should not be inserted silently into a report whose label tells the reader it describes the proportion of records meeting a common criterion.
I also ask whether the regional populations and criteria are sufficiently comparable before combining them. In this fictional case, the criterion was shared. In other work, a different definition or observation period could create another problem even if the arithmetic used the counts correctly. Weighting is not a substitute for understanding what each record means. A pooled percentage becomes useful only when its underlying account is coherent enough for the intended comparison.
The practical implication is that the report needs a short, intelligible explanation of its population. Executives should not require a statistical lecture every time they read a rate. They should be able to find what was counted, what qualified and how the combined account was formed. Those details make challenge possible when the result seems surprising, and they help prevent a summary from drifting into a claim it was never designed to support.
The correction needs its own accurate explanation
Moving the reported figure from 70% to 54% could itself have created a misleading story. The difference was sixteen percentage points, but it was not evidence that the underlying records had deteriorated between two dates. We had changed the aggregation to answer the stated question correctly. The explanation needed to separate a repair to the measurement from a change in the measured work.
That separation was important for the people whose activity appeared in the report. A poorly explained correction could make one region seem to have caused a sudden decline or suggest that management had uncovered a new operating failure. The reviewed evidence did not establish either claim. It showed that the earlier combined presentation gave equal influence to unequal populations. I wanted colleagues to understand that without minimising the responsibility for having approved it.
The ₹19 lakh cost also stayed in the account as report repair. I did not turn the corrected percentage into a financial benefit or claim a subsequent improvement in completeness that had not been measured. A more accurate summary can support better decisions, but the case does not automatically prove those later decisions or their outcomes. A leadership narrative should stop where its evidence stops, even when the intervention feels obviously useful.
The incident changed my review habit. I now ask what survives and what changes when several results become one number. Does the summary preserve the unit the reader expects? Are differences in population size visible in the calculation? Could another legitimate question produce a different summary, and if so, which one are we answering? These are business questions expressed through data, not checks that can be delegated entirely to a charting tool.
I want that habit to be constructive. The answer is not to make every report longer or to refuse any simplification. Leaders need summaries precisely because the work is complex. The aim is to simplify without changing the question unnoticed. In this case, keeping the underlying counts in view would have made the combined result both shorter to defend and more faithful to the records. That is the standard I now bring to a percentage before asking someone else to rely on it.