Chief Data Officer / India
The number changed meaning before the decision
Decisions & consequences
01Keeping 110 unanswered quotations out of a false conclusion
In 2023 a capacity paper used a quotation-conversion figure to support a proposed ₹17.5 crore commitment. Our cohort contained 350 offers: 98 won, 142 lost and 110 still unresolved at the review date. The dashboard showed 28% using every offer as its denominator. A second presentation showed 40.8% among resolved offers. Neither calculation was an error, but both were being used as though they described the eventual outcome of the entire cohort.
Dev Drelzavi, our commercial director, needed a planning assumption. His argument for using the resolved population was practical: it contained real decisions, while the pending offers might take months. I argued that the time to decision was itself related to the type of purchase. Repeat requirements and new applications did not mature at the same pace. Selecting only completed decisions could therefore change the mix of opportunities being used to justify capacity. Counting every pending offer as a loss introduced the opposite problem.
We spent ₹29 lakh on a defined review of quotation dates, decision status and customer follow-up. The report retained both descriptive percentages, showed the 110 pending offers separately and grouped offers by time since issue and commercial type. It did not invent final outcomes to complete the chart. I recommended that management examine a range of demand assumptions rather than adopt either snapshot as a settled conversion rate.
The capacity proposal was deferred four months while commercial and operating colleagues revisited its basis. That is not ₹17.5 crore saved: the investment remained an option, and the delay had an opportunity cost. Nor did the review prove that pending offers would convert at a higher rate. Its result was a more honest account of the decision still to be made. I should have surfaced the unfinished cohort before the dashboard reached the investment discussion, rather than waiting until two apparently conflicting percentages forced the issue.
Refusing to turn sixteen catchments into hundreds of experiments
A 2024 distributor-offer trial allocated sixteen commercial catchments to two routes, eight in each group. The alternative offer combined a different reorder communication and service arrangement; price and technical suitability remained subject to normal commercial approval. The experiment cost ₹54 lakh. At the end of the agreed observation period the offer group showed 8.6% higher aggregate sales than the comparison group, and colleagues wanted that figure to support a broad rollout.
The commercial case was understandable. Orders had arrived, the trial had cost money and distributors wanted a decision. However, the assignment had occurred at catchment level. Hundreds of transactions did not create hundreds of independent assignments. Eleven buyers also traded across the routes during the trial. Removing those buyers after observing their behaviour would have changed the comparison in a way the original design had not promised. My earlier approval of the trial had not given enough attention to that connected trading pattern.
We retained the assigned-group account, documented the cross-route activity and examined how sensitive the aggregate result was to individual catchments and large orders. The analysis treated the 8.6% as an observed difference, not a proven effect attributable solely to the offer. I recommended against immediate broad rollout. The alternative was to proceed on the promising commercial signal and monitor results, which would have been a management choice under uncertainty rather than an evidence-established expansion.
Meera accepted a redesign of the next trial instead. We had spent ₹54 lakh without obtaining the decisive answer the original paper implied, and I recorded that limitation alongside the useful operating learning. The result did not prove the offer ineffective. It showed that our design and network gave us less independent evidence than the presentation suggested. My lesson was to map how buyers cross commercial boundaries before choosing the unit of assignment, and to agree the decision standard before the sales chart makes one conclusion emotionally attractive.
Separating purchasing inflation from a changed material basket
A 2025 purchasing review showed average invoice cost per kilogram rising from ₹146 to ₹159 across the selected input population. A draft management paper treated the full increase as price pressure and included a ₹2.1 crore procurement-leakage hypothesis. The procurement head disputed the interpretation, pointing to a different mix of material specifications. Finance needed an explanation that could inform the forecast without simply accepting either team's preferred account.
My team undertook an ₹18 lakh analysis with procurement and finance. We defined comparable units and terms, checked the treatment of freight and taxes, and constructed a matched fixed basket where specifications could be compared. On that basket the corresponding measure moved from ₹146 to ₹150 per kilogram. The difference between that movement and the observed invoice average involved composition and interaction effects; it was not a clean residual we could confidently label as one cause. Some unmatched purchases also needed separate interpretation.
The competing approach was to retain the broad average as a quick warning measure and investigate exceptions later. I supported retaining it as a description of what the company had bought, but opposed using it alone as a judgement about supplier price behaviour or procurement performance. The fixed basket had limits too: it represented a defined comparison, not every material the business might require in the future. We presented the two measures together with their scope.
The ₹2.1 crore leakage hypothesis was withdrawn from the draft working account pending a better basis. No cash was recovered by withdrawing a hypothesis, and the ₹18 lakh study remained an expense. Finance retained responsibility for the actual forecast and accounting treatment. The work changed which questions were asked of procurement: which specifications had shifted, what drove the buying mix and where comparable prices had genuinely moved. I would bring procurement into the definition of comparable items earlier next time, before a simple average starts to sound like a verdict on its decisions.
Restoring the date the customer was first promised
In a 2026 service-report review we examined 420 completed orders from a defined period. The current report showed 382 delivered on time against their latest recorded date, or 91.0%. Against the original committed date, 330 were on time, or 78.6%. Eighty-six orders had a changed date; for fifty-two, the revision changed the on-time classification. The remaining thirty-four revisions did not change that classification. The report had been updating its reference date without making that shift clear to readers.
Operations colleagues explained that some revisions reflected genuine customer changes and agreed commercial adjustments. They were right that the first date could become an unfair measure of the revised undertaking. Commercial colleagues nevertheless wanted the first promise visible, because it described an important part of the customer's experience. I rejected the suggestion that one percentage had to replace the other. Each answered a different question, and neither could explain the reason for a date change on its own.
We spent ₹36 lakh reviewing the date history and revising the report. The two measures now appear with explicit labels, the revision population and reasons available for operational interpretation. The change retained both the initial commitment and the later agreement instead of allowing one field to erase the earlier account. We did not use the exercise to rank employees or infer deliberate manipulation from a changed date; the underlying circumstances required proper review by the responsible managers.
The outcome was a clearer service discussion, not an immediate improvement in deliveries. No order arrived sooner because its history was restored. We could now ask whether a customer-requested change, an unrealistic initial commitment or an internal delay explained a revision. My responsibility was to make that distinction possible. I should have required the report specification to state which promise it measured when the metric was first approved, rather than assuming that the familiar phrase on time had one stable meaning across commercial and operating teams.