One order can appear in several campaign reports

The proposal asked for another ₹1.6 crore of campaign expenditure. Its supporting page added order counts from three platforms, and the total looked substantial: 18,400. Our online order record contained 12,000 distinct orders in the same reviewed period. The arithmetic on the page was correct. The interpretation was not. We had added claims made under different attribution systems as if they described separate purchases. Before asking the chief executive to consider the additional investment, I needed to repair the account on which the recommendation rested.

This was not a case in which one platform had necessarily supplied an invented number. Each report answered a question within its own system. A purchase could fall within more than one attribution window and have a relationship with more than one recorded interaction. The problem arose when we combined those reports without preserving their meaning. As CMO, I was responsible for the recommendation that used them. Pointing to the agencies or the platforms would not have answered why our own management paper treated overlapping claims as a business total.

## Reconcile the counted object before discussing efficiency

The three reports contained 8,300, 6,200 and 3,900 attributed orders. Their sum was 18,400. We held the proposed incremental investment while examining how those records related to the 12,000 distinct orders. The ₹1.6 crore was a proposal, not an approved expense that we later recovered. I therefore did not report it as money saved. It was a decision postponed because the current case was insufficient, with an opportunity cost as well as a possible benefit. Some worthwhile activity might be delayed while we clarified the evidence.

The measurement review cost ₹29 lakh. It examined the relationship between the reports and the period's order records. We identified 9,600 distinct orders with one or more attribution records and 2,400 without one in the reviewed set. Together they accounted for the 12,000 orders. Those categories were useful because they stopped us counting the same purchase repeatedly in an enterprise total. They did not reveal every influence on the buyer, and the second category did not establish that marketing had no role in those purchases.

I wanted the reconciliation explained in ordinary language before it appeared as a more sophisticated chart. The unit was a distinct order within the agreed period and scope. An attribution record was a claim about a relationship under a platform's rules. Those are different objects. A report can count both, provided the reader knows which is which. The failure was to let the familiar word order travel from one context to the other without its qualification. Adding a footnote after an investment recommendation would not fully repair that misunderstanding.

The agencies still needed their platform views to manage activity within those environments. Removing those reports entirely would have made some daily decisions harder without creating better evidence. We kept them for their proper purposes and stopped using their sum as an enterprise sales result. This distinction helped the conversation remain practical. We were not declaring every platform metric worthless. We were deciding which questions each source could help answer and where a separate business-level account was needed.

## An attributed sale is not an incremental sale

Deduplication is a counting repair. Incrementality is a causal question. Knowing that 9,600 distinct orders had at least one attribution record does not tell us how many would have occurred without the campaigns. Some buyers may already have intended to purchase. Some may have encountered several messages. Others may have been influenced in ways the selected records did not capture. A cleaner count should not become a more confidently stated claim about causation unless the evidence actually supports that move.

This mattered particularly because the proposed investment was additional. Management needed to consider what another ₹1.6 crore might achieve, not merely whether existing activity had appeared near purchases. The marginal decision could be different from a judgement about the whole programme. A channel might be useful at its current level while offering poor evidence for expansion. Conversely, weak attribution data might coexist with a commercially important activity that needs a different measurement approach. The corrected report did not settle either possibility by itself.

I discussed the limits with finance because the language of marketing efficiency often crosses into the language of return. Spend divided by an attributed count can be a descriptive ratio, but the denominator needs a precise definition. If the count overlaps across channels or does not establish incremental orders, the resulting ratio cannot quietly become a cost of acquiring additional business. A director should be able to see the assumptions without reconstructing several agency reports. That is part of the CMO's work in presenting a recommendation.

We also needed to avoid making the correction look like a hidden admission that the entire past programme had failed. The available evidence did not establish that conclusion. It established that a particular combined report could not support the claim we had placed on it. I wanted colleagues to be willing to surface such problems early. If every correction is treated as proof of total failure, people have an incentive to defend a number long after its meaning becomes doubtful. A useful review can be demanding about the claim and fair about what the finding does not show.

## Make the next investment answer a narrower question

The next proposal needed a defined question, a suitable comparison and a clear account of the limitations. Depending on the activity, that might require a structured test or another method developed with the relevant specialists. I would not prescribe an experiment merely because the term sounds more rigorous. Retail settings contain spillovers, timing effects, uneven markets and operational constraints. The method needs to fit the decision, and its cost needs to be considered alongside the proposed commercial exposure. Precision in the question comes before sophistication in the method.

We had not established a later experimental return by the time of the review described here. I do not attach a positive ROI figure to complete the story. The outcome was a reconciled order account, a stopped practice of adding overlapping platform claims and an investment proposal that needed new evidence. That is a less dramatic ending than a campaign rescue. It is also a more accurate account of the decision. A leadership profile should be able to show a useful correction without manufacturing the eventual commercial result.

For a board discussion, I would separate the historical description from the forward choice. The first explains what happened in the reviewed period and how confidently we know it. The second explains the proposed expenditure, expected benefit, downside and learning that would justify proceeding. Mixing the two encourages a familiar count to stand in for a forecast. Keeping them distinct makes it easier for directors to challenge the right assumption and for management to explain what would change its recommendation.

My own lesson was that measurement language belongs in the creative and commercial conversation early. It cannot be delegated entirely to the final reporting stage. If a campaign is sold internally with a promise that its chosen data cannot test, the later team inherits an avoidable problem. I now want the proposed claim and the available evidence discussed together. One purchase may legitimately appear in several reports. Leadership begins by resisting the temptation to make those appearances add up to several purchases, and continues by asking what the repaired account still cannot tell us.