A falling total can contain two improving service rates

The combined completion rate had fallen, and the first draft said the operating teams had deteriorated. Area managers objected because each of their reported rates had improved. The disagreement appeared to concern which numbers were correct. It actually concerned what population each number described. We had changed the mix of scheduled visits in the observation account. The overall percentage and the two area percentages were all arithmetically valid, yet they moved in different directions. Choosing the account that suited a preferred story would have concealed the decision problem rather than resolve it.

The first period contained 1,000 scheduled core-area point-visits, with 900 recorded completions, and 200 extension-area point-visits, with 100 completions. The second contained 500 core-area visits with 460 completions and 600 extension-area visits with 360 completions. Core-area completion therefore rose from 90% to 92%, while extension-area completion rose from 50% to 60%. Combining the groups produced 1,000 completions from 1,200 visits in the first period and 820 from 1,100 in the second. The corresponding pooled rates were approximately 83.3% and 74.5%.

The arithmetic was not the disagreement

The second account gave much more weight to extension-area visits, where recorded completion remained lower despite its improvement. That change in composition helped explain why the combined direction differed from the directions within the two groups. It did not establish why the schedules changed or why either area's completion rate moved. Those were further questions. My responsibility was to stop the report from using one level of the account to make an unsupported statement about another. We needed a view that retained the population weights as well as the percentages people wanted to compare.

The reporting sponsor reasonably wanted one concise measure of the combined service. Management could not ignore the fact that a smaller proportion of the scheduled visits in the second account had recorded completion. The area managers also had a valid point: the pooled decline did not prove that each group had worsened. Neither side needed its evidence removed. The task was to explain how the accounts related and which question each could support. A combined measure can describe the overall observed service while remaining inadequate as an unqualified judgement about the performance of its constituent teams.

Definitions mattered before any interpretation of the trend. These were scheduled point-visits, which can include repeated visits to a location. They were not unique households and did not establish household coverage across a town. Recorded completion depended on the evidence and status rules used in the account. The labels core and extension identified the groups used for this review; they did not assert that all locations within a group had identical conditions. If those distinctions disappeared, the report could remain numerically accurate while encouraging several conclusions its observation design never addressed.

Choose the question before choosing the headline

A useful review begins by asking what management needs to understand. Is it the combined recorded service in the two periods, the comparison within defined areas, the reasons the scheduling mix changed or the effect of a particular intervention? Those are related questions, but they require different evidence and interpretation. I wanted the sponsor to choose the question openly rather than let the most convenient available percentage choose it by default. A dashboard should not force every discussion into a single arrow merely because that is the easiest visual element to place beside an executive summary.

We spent ₹34 lakh retaining the four numerator-denominator pairs, their area descriptions and the combined account in the reporting process. We removed the claim that the pooled decline alone established team deterioration. We did not remove the decline itself. Operations retained responsibility for investigating schedules, practical constraints and the reasons behind recorded completions. The data work gave that discussion a more accurate starting point; it did not make the data function the authority on how collection should be organised or which worker was responsible for a particular outcome.

One tempting response would have been to select a fixed population mix and publish a more favourable adjusted rate. Such a calculation can answer a defined comparison question if its assumptions and weights are explicit. It would not be the actual pooled rate observed in either period. We did not introduce an adjusted figure here and allow it to stand in for the operating account. The immediate requirement was to preserve what had been observed and clarify the question. Further analytical treatment would need a reason tied to the decision, rather than serve as a way to rescue the story management preferred.

A clearer comparison still needs operating judgement

The extension-area rate rose from 50% to 60%, but it remained below the core-area rate. Showing improvement should not make the remaining gap disappear. Equally, the lower level did not by itself establish why the gap existed or what action would address it. Service conditions, schedules, recording and other factors required examination by the responsible owners. I wanted management to hold two ideas at once: the group had improved within this account, and its recorded completion still raised a practical question. Good scrutiny does not require choosing between acknowledgement and challenge.

The revised report made the discussion less convenient. It demanded more than a quick agreement that the company was improving or deteriorating. Managers had to explain the shift in observed work and examine the remaining differences without using the aggregate as a shortcut. That extra effort was part of the decision's cost. The result was not a demonstrated improvement in service, a causal evaluation of teams or a new estimate of household access. It was an account in which the relevant denominators remained attached to the claim, allowing the next investigation to begin from a more precise question.

This experience changed how I review recurring performance summaries. I look for changes in the population before interpreting changes in the rate. A growing extension programme, a different mix of locations or a revised observation schedule can alter what a total contains. That does not make the total unimportant or excuse poor outcomes. It means the explanation should follow the actual composition of the work. If we want a comparison of like conditions, we need to define those conditions and explain the limits of that comparison, rather than assume that two percentages with the same label necessarily measure the same task.

For directors, the practical test is whether management can show the people, activities or locations behind a headline. A board need not conduct the statistical work itself to ask why the composition changed and what the proposed interpretation leaves out. It should also resist a switch to whichever level looks most reassuring. In our review, the lower combined rate and the higher component rates belonged in the same conversation. Keeping them together did not remove judgement. It made the judgement visible, with operating causes still to investigate and the observed service still available for scrutiny.