Let the learning promise stay close to what was observed

In 2022 I approved a campaign headline describing 40% fewer mistakes after a customer-service simulation. The underlying fictional review followed 80 learners whose average errors in a defined scenario fell from 12 to 7.2. That arithmetic supported a 40% relative change within the exercise. It did not establish a reduction in workplace mistakes. We had no comparison group and no follow-up observation of participants doing their jobs. My headline had removed those conditions and left a buyer free to read a stronger conclusion.

The campaign team argued that the result was still impressive and that adding limitations would make the material cumbersome. Meera Vaqsumi, our learning-evaluation head, asked which outcome the headline actually described. I could not defend it by pointing to correct subtraction. The problem was the move from a practice observation to a claim about work, and the suggestion that the product alone explained the change. I withdrew the broad statement and authorised revision of 14 assets at ₹18 lakh cost.

We retained the scenario result with its population and measurement context. We also made clear that repeated practice, familiarity with the exercise and other influences had not been separated through the review. The correction did not say the software lacked value or that the learners had learned nothing. It said the available evidence answered a narrower question. A prospective customer still needed to consider how it would assess application in its own workplace.

I gave up a simple headline after publication and accepted the cost of correcting work my team had already completed. Sales colleagues had to revisit material they had begun using. I did not attach a later conversion improvement to make the decision look commercially effortless. The result was a more accurate account and an unresolved evaluation question, not a demonstrated increase in sales or a measured workplace benefit.

The case changed my approval practice. I now ask the campaign owner to state the observed setting and the intended claim side by side before the copy is shortened. Marketing remains responsible for the meaning a reasonable reader can take from the final words, even when the calculation beneath them is correct.

A 2023 survey invited 400 participants to comment on a software-supported learning programme. One hundred replied, and 85 of those responses were positive under the survey's defined question. The campaign draft described 85% approval without keeping the respondent population visible. In a presentation about the full programme, the wording invited the audience to treat the percentage as though it described all 400 participants. Three hundred people had supplied no answer.

The account team wanted to publish while the programme was still recent. They argued that every survey has nonresponse and that the available replies were encouraging. I agreed that the replies were useful, but they did not make the silent group disappear. Meera Vaqsumi asked us to distinguish the satisfaction question from both learning and workplace application. Even a fully answered satisfaction survey would not by itself establish either of those outcomes.

I stopped the campaign statement and kept the result as 85 positive replies among 100 respondents from 400 invitations. We did not recode the 300 nonresponses as negative, neutral or positive. I approved ₹23 lakh for further research intended to examine the missing perspectives and the questions the first survey had not answered. That work remains unfinished in this account. Its budget is a commitment to investigate, not evidence that the initial result has since been confirmed.

The decision cost the team a timely promotional claim and required a less satisfying conversation with colleagues who wanted a single score. It also prevented us from solving uncertainty through a convenient assumption. We could explain what respondents had said while acknowledging that the broader population was not represented by the same evidence. No subsequent purchase or retention result is attributed to this correction.

I would make the same distinction in a board paper. Directors can ask whether the people behind a percentage match the population named in the conclusion. They do not need to design the questionnaire to notice that a large unknown group has vanished between the collection of responses and the commercial statement. Preserving that gap makes a research decision possible; hiding it makes the question look settled.

A 2024 demonstration programme recorded 360 sessions. The first expansion proposal treated that number as evidence of a broad audience and sought ₹1.7 crore for a larger programme. Reviewing participant records showed that 90 people had each taken part four times. The activity count was accurate, but it represented repeated participation rather than 360 distinct people. Fifty-four of the 90 requested follow-up and 36 did not.

The event team argued that repetition was evidence of interest and should strengthen the investment case. I accepted the first part of that argument. A person choosing to return can be commercially meaningful, and the programme had created opportunities for deeper practice. The proposed expansion, however, was being justified partly by reaching more distinct people. Repetition could not answer that question simply because it was valuable for another purpose.

I cancelled the ₹1.7 crore expansion pending a clearer case about distinct reach. The ₹31 lakh pilot expense remained spent. We retained the 90-person participation account and the 54 follow-up requests, without presenting those requests as purchases or attributing future contracts to the event. The 36 who did not request follow-up were not labelled uninterested; the record only established that they had not made that request through the programme.

This was a difficult decision because it came after a lively demonstration that colleagues felt had worked. I did not ask them to describe the pilot as a failure. I asked them to separate what it had demonstrated from the specific reason given for spending more. A smaller, more accurate account could support a different proposal later, but that proposal would need its own evidence and approval. The cancelled amount was not reported as a realised operating saving.

The case taught me to examine the noun beside a number before debating its size. A session, a person and an enquiry belong to different stages of the commercial story. When those nouns change silently, a sound activity record can become a misleading investment argument. Marketing should preserve the distinctions while still recognising the value of repeated engagement.

In 2025 a customer-facing pack described 6000 active users of a software deployment. The underlying count was 6000 provisioned accounts. Of those, 4200 had been activated and 1800 had never been activated. A defined monthly-active measure recorded 2100 accounts, a subset of the 4200 activated population. The pack had taken the broadest available count and attached a label that suggested current use.

The customer team worried that correcting the wording would make the deployment look smaller after a successful provisioning effort. Product colleagues pointed out that activation and monthly activity answered different questions and depended on the dates and definitions used. I supported revising 16 collateral items at ₹12 lakh remediation cost. The materials now distinguished provisioned, activated and monthly-active accounts rather than using a single flattering number.

We did not infer that every unactivated account represented a failed implementation. Some accounts might have been created ahead of a later programme, but that explanation would need evidence from the responsible customer team. Nor did monthly activity prove learning, meaningful practice or workplace application. The correction created a clearer set of questions about adoption instead of resolving them by choosing a stronger adjective.

I accepted that the revised account would require more explanation from sales and marketing. It removed an easy headline and exposed a stage of the deployment that still needed attention. The change did not alter a contract amount, recognised revenue or annual recurring revenue, and this case makes no claim that those financial measures improved afterwards. Our responsibility was to describe the records accurately within marketing material and refer operational questions to their owners.

The practical lesson was to maintain a dated definition beside every adoption measure. A large account population can be important for capacity and rollout planning, but it should not be borrowed as evidence of use. Once the distinction was visible, we could discuss the customer journey more honestly: who had access, who had begun and who appeared in the specified month's activity record. Those are connected questions, not interchangeable answers.

A 2025 showcase preparation involved 120 participants, including 102 using a desktop interaction path and 18 using keyboard-only paths. The defined review found that 12 of the 18 keyboard-only paths could be completed while six were blocked by a drag control. The planned presentation described an inclusive practice experience, but the reviewed interaction prevented part of that audience from completing the exercise.

The campaign team proposed showing the smooth desktop sequence while product colleagues worked on the alternative. They argued that the public date was important and that the demonstration could still communicate the intended experience. I thought that would make the marketing account depend on the very exclusion the review had exposed. I supported postponing the showcase for six weeks and allocating ₹27 lakh to the defined rework and recheck.

The team revised the interaction and reran the 18 reviewed keyboard-only paths. All 18 were completed in that recheck. We reported the correction at that scope. It did not establish whole-product accessibility, compliance with every requirement or successful use by people with every disability. Qualified evaluation and broader testing retained their own place. The result was evidence about a specific barrier and the paths examined after the change.

The postponement disrupted a prepared campaign and required colleagues to explain why the public date had moved. I did not turn the delay into a claim about sales protected or reputation saved. The known consequence was six weeks of postponement and ₹27 lakh cost, with the reviewed barrier corrected. The original 120-person preparation group also remained distinct from a universal statement about customers.

This decision made inclusion part of the release conversation rather than a descriptive line added afterwards. Marketing can ask whether the experience it promises is available under the conditions shown, and it can support a delay when the answer is incomplete. It should also resist turning one successful correction into a larger claim than the recheck supports. The value of the work is clearer when its boundary remains visible.