When the direction of a total tells only part of the story

## Begin with the counts rather than the arrow

A falling percentage invites a familiar management conversation. Someone asks what deteriorated, another person offers an explanation and a proposed response follows. The direction of the number can become the premise before anyone examines what population produced it. That sequence is understandable when a board receives many measures and limited time. It is also dangerous when a changing composition can move the total in a different direction from every relevant group inside it. The first useful question is not always why the rate fell. It may be which counts were combined to produce it.

I learned this through a fictional callback report I approved. In the first period, digital requests completed within their promised slot numbered 800 out of 1,000, or 80%. Assisted requests numbered 180 out of 300, or 60%. The combined rate was 980 out of 1,300, approximately 75.4%. In the second period, digital requests were 450 out of 500, or 90%, while assisted requests were 650 out of 1,000, or 65%. Both conditional rates had risen. The combined rate was now 1,100 out of 1,500, approximately 73.3%.

The arithmetic did not contradict itself. The second period contained a much larger share of assisted requests, whose rate remained lower than the digital rate. The total described that new mixture. My error was the heading that treated its decline as a sufficient description of the service's deterioration. I had checked reconciliation and approved an interpretation that needed further examination. The fact that the numbers added up did not establish that the sentence above them was fair. A data leader is responsible for that sentence as well as for the calculation beneath it.

The remedy began by keeping the counts visible. Percentages alone hide the size of the groups that give them their influence. A reader can see 80% becoming 90% and 60% becoming 65% without understanding why the total behaves differently unless the denominators are present. The counts also help identify whether the categories themselves have changed meaning or whether a comparison spans different operating circumstances. Those questions require the people who understand the workflow. A statistical explanation should open that discussion, not substitute for an operational account of the business.

## Retain the combined experience as well as its composition

Once the group results are visible, another tempting simplification appears: declare the aggregate misleading and replace it with the more favourable conditional story. That would repeat the same mistake from the opposite direction. The overall measure still described the observed population, assuming the stated definitions were applied consistently. It mattered to the company and to the people using its service. A changing mix did not make their combined experience imaginary. The purpose of disaggregation was to enrich the account, not remove whichever measure was less comfortable for management.

In our example, the first period contained 320 missed promises and the second 400. Those counts deserved to remain beside the improved conditional rates. A business can face more unmet commitments even while each channel's proportion improves, because its volumes and composition have changed. The board might need to consider capacity, access, demand and the suitability of the service model. It should not be encouraged to believe that an explanation of the total has resolved those operating questions. Explaining why a measure moved is different from deciding that the experience is acceptable.

The service sponsor preferred one headline because it was concise and corresponded to the actual combined result. That objection was useful. It reminded us that a more detailed account can become harder to use, and that management cannot examine every possible subdivision in every meeting. We spent ₹52 lakh on reconstruction and reporting work, retaining the aggregate and two meaningful channel views. The aim was not to produce an unlimited collection of cuts. It was to show the composition necessary to assess the specific judgement the original heading invited.

Choosing categories therefore needs discipline. A group should have operational meaning, a stable enough definition and a reason to belong in the comparison. Searching many subdivisions until one produces a favourable result would not repair the original account. Nor should a convenient channel label conceal meaningful variation within it. I want the paper to explain why these groups are useful for this question and what remains outside the analysis. That gives directors a basis for asking for more detail when necessary without treating every management measure as an invitation to reconstruct the entire database.

## Separate a better explanation from a causal conclusion

The revised account did not establish what caused either conditional rate to improve. Staffing, demand, request complexity, the definition of a promised slot and other circumstances could matter. The two periods were observations, not an experiment designed to isolate an intervention. A clearer relationship between counts and percentages cannot supply evidence that was never collected. I wanted the paper to resist attaching a programme success claim to the revised result merely because we had withdrawn the earlier deterioration claim. The appropriate conclusion was more qualified than either celebration or blame.

This distinction affects the division of responsibility. Operations remained accountable for the callbacks and for examining the service conditions. My function was accountable for the measure's construction, explanation and limits. Underwriting, finance and other professional owners retained their own responsibilities. A CDO can insist that a number be interpreted carefully without becoming the manager of every activity it measures. For a board, the useful arrangement is one in which the analytical and operating accounts can be questioned together, while the decision rights remain clear enough for follow-through.

I would change the sequence of review next time. The composition question should come before the proposed management heading is approved, especially when the measure may influence resource allocation or a judgement about someone's performance. Retrospective correction is necessary when an error is found, but it carries the cost of withdrawing a confident statement already circulated. The original wording should stay in the learning record with the reason it changed. Quietly replacing the heading would make it harder to see that the failure concerned interpretation rather than arithmetic.

For my own leadership profile, this is a more useful example than a claim that I make data-driven decisions. It shows a particular number, a mistaken implication I accepted, a colleague's challenge and a revised account that still leaves work for management. An executive website should make that chain understandable without asking the reader to become a statistician. The lesson is modest but consequential: before a directional measure becomes an explanation, inspect the population that gives it direction. Then keep the people behind the inconvenient result visible while deciding what the business should do next.