Chief Data Officer / India
Keep the evidence and narrow the inference
- Data organisation
- 29 colleagues
- Four heads coordinating observation, research and publication data
- Annual function expense
- ₹5.8 crore
- Includes ₹1.6 crore methodology development
- Research and data career
- 33 years
- Source records through commercial interpretation and publication accountability
Decisions & consequences
01I withdrew a sales conclusion built from removal records
In 2018 I approved a selected account that treated thirty-six removed vehicle listings as completed sales. The removal records were accurate, but the conclusion went beyond what those records established. Farhan’s review found supporting transaction evidence for twelve cases. The outcome of the other twenty-four removals remained unknown.
The distinction mattered because a customer could read the original account as evidence of completed market activity. A listing could disappear for reasons that our selected source did not establish. I had accepted the convenient interpretation rather than preserving that uncertainty. The correction therefore belonged to the publication decision I owned, not to a source provider supposedly failing to record the event we wanted to infer.
We spent ₹33 lakh over eight weeks correcting the account and reviewing the basis for approval. The replacement statement separated twelve confirmed transactions from twenty-four unknown outcomes. We did not label those unknown cases unsold, infer a sale date or invent a transaction amount. The original removal observations remained useful as source events, provided they were no longer presented as a complete sales account.
I explained the narrower result to commercial colleagues and kept Farhan’s challenge in the decision record. The work did not establish a wider-market completion rate or a customer benefit that had not been measured. It established that our selected publication needed to change, and that a precise count could still be attached to the wrong commercial meaning.
The lesson for my later CDO role was to ask which event a sentence claims to describe. A reader should not need to reconstruct the source process to discover that the headline has crossed from observation into inference. Unknown outcomes deserve a visible place in the account when the evidence cannot responsibly assign them elsewhere. The correction was more useful because it stopped at that boundary.
A changed vehicle mix explained the selected mean difference
In 2023 two selected groups of twenty transactions appeared to show a twenty per cent increase in average price. The first group contained fifteen vehicles in an older band averaging ₹8 lakh and five in a newer band averaging ₹16 lakh. Its total was ₹200 lakh and its mean ₹10 lakh. The second contained ten in each band at the same respective means, giving ₹240 lakh and an overall mean of ₹12 lakh.
The arithmetic was correct. The problem was the proposed interpretation of the change as a general price movement. The proportion of newer-band vehicles had increased. That composition difference explained the mean change in this constructed account without requiring a change in either band’s average. It did not establish that every vehicle was comparable or that individual prices had remained unchanged.
Farhan and the methods colleagues examined the account with my team. We spent ₹37 lakh over nine weeks revising how the comparison was described and shown. I asked the publication to retain the two group sizes, the band proportions and their means so that a reader could understand the calculation. The narrower explanation could still be commercially useful without being presented as a complete constant-quality index.
Age was only one characteristic. Vehicle class, condition and transaction circumstances could still differ within the bands. We therefore did not turn this example into a claim that there was no wider-market inflation, or into a valuation method for a particular vehicle. The result was a corrected interpretation of selected fictional groups, not an estimate of actual Indian market prices.
My decision was to make the source of the difference visible before allowing the headline to describe it. A familiar percentage can feel more conclusive than the comparison beneath it. The CDO contribution was to preserve the composition question in the customer-facing explanation and support the method owner when that question narrowed the commercial claim.
The bundle remained evidence without becoming twelve observed prices
A selected 2025 account contained eighteen deals: sixteen single-vehicle transactions and two bundled transactions. One bundle covered twelve vehicles for a combined ₹1.44 crore. An equal division produced ₹12 lakh per vehicle. That calculation was straightforward, but it did not establish that each of those twelve vehicles had achieved the same individual price.
The vehicles differed in their characteristics, and the transaction evidence described the combined deal. A single-vehicle benchmark needs a treatment that respects the unit actually priced. We could not make the account more complete by silently converting an allocation into twelve independently observed sale prices. At the same time, simply hiding the bundled deal would remove relevant commercial activity from the broader research story.
I worked with Farhan to separate the bundled transactions from the single-vehicle comparison and retain them in a clearly described combined-deal account. The review cost ₹54 lakh over ten weeks. Sixteen single-vehicle deals remained in the selected individual comparison; the two bundles were disclosed separately. The twelve-vehicle bundle’s total remained visible without assigning an unsupported achieved amount to each vehicle.
The result was a more explicit publication scope. It did not establish that the single-vehicle set represented every market segment, that the allocated average was commercially meaningless, or that a particular vehicle should be bought or sold at any stated figure. An allocation can serve a planning purpose if it is named as such. It cannot acquire the status of an observation merely because division produces a precise answer.
I wanted the reader to know both what we retained and what we excluded from the comparison. A methodological choice changes the coverage of a benchmark. The responsible account explains that consequence rather than presenting the remaining table as if the bundled activity had never existed. That balance between visibility and comparability became the main leadership decision in the review.