Chief Data Officer / India
A clearer account of who answered and what could be compared
- Data organisation
- 40 colleagues
- Five leads across research data and product definitions
- Annual function expense
- ₹7.8 crore
- Includes ₹2.2 crore improvement work
- Survey and data career
- 30 years
- Survey support through audience-research data leadership
Decisions & consequences
01Correcting a pooled-interview claim I had approved as distinct people
In 2019 I approved a client report describing a pooled file of 2,400 interview records as 2,400 different people. The file combined repeated research waves. Some respondents had taken part more than once, and the selected pool contained 1,650 distinct respondents. The source records were not invented within the study account; the problem was the unit my approved wording asked the client to believe had been counted.
The commercial team favoured retaining the larger number because it was familiar and reflected completed research activity. Zoya argued that activity was not the same as a count of different people. I accepted that challenge and reopened the account. We spent ₹39 lakh over eight weeks correcting the report and the approval process that moved the pooled total into client-facing language. Both numbers remained useful when labelled correctly: 2,400 interviews and 1,650 distinct respondents in the selected pool.
We did not describe the correction as a new estimate of population reach. Nor did identifying distinct respondents establish that all observations were independent or that the sample represented a wider population. Those questions required their own methodological treatment. My responsibility was to stop an incorrect unit from becoming the basis for further commercial inference. The revised account preserved repeated participation rather than treating it as an inconvenience to erase from the file.
The trade-off was giving up a larger headline and revisiting a report already approved by the data function. I kept my own approval in the story instead of presenting the issue as a problem created by field colleagues. The bounded result was a corrected claim and a clearer review of the relationship between respondent and interview. There was no measured sales or confidence gain to add. The case changed how I examine pooled products: before asking whether a number is impressive, I ask which entity it counts and whether that meaning survives the move from source records to the chart a client will remember.
Breaking the apparent trend when interview collection moved online
In 2023 a continuing study moved from telephone interviews to web collection after eight telephone waves. The next chart could easily have appeared as the ninth point in one uninterrupted trend. The commercial attraction was continuity: clients already knew the series and wanted to compare the new result with earlier reports. The methodological question was whether a change in how people answered could affect the comparison.
We examined a paired comparison involving 120 people. Ninety gave the same answer and thirty gave a different answer on a selected item. That was a reason to investigate comparability, not proof that the collection method caused every difference. Timing, response variation and the comparison design also mattered. I worked with Zoya to keep those limits explicit rather than use the thirty differences as a universal mode-effect estimate or an adjustment to apply across the entire product.
We used ₹58 lakh over twelve weeks to revise the data product and its explanation. The web wave began a separately labelled series while the earlier telephone results remained available with their method identified. We did not discard the historical work or claim that no future comparison could ever be supported. We stopped the current presentation from implying direct comparability before that case had been established through appropriate further work.
The cost was a less convenient client story. A continuous line would have been easier to show, but could have encouraged a business decision based on apparent movement that the study design had not yet made interpretable. The result was a product whose method change remained visible. No quantified improvement in accuracy or client decisions is claimed. I wanted the commercial team to explain the value of the new wave without borrowing certainty from the old series, and the data team to recognise that a familiar chart structure is itself a claim about what can be compared.
Removing district coverage that demographic weighting could not create
In 2025 an online audience panel recruited across eighteen towns was being marketed as coverage of thirty-two districts. The proposition relied partly on demographic weighting to suggest that a broader geographic claim could be made from the observed respondents. Weighting had a defined analytical purpose, but it did not create observations in locations outside the recruitment frame or settle the selection limitations of an online panel.
The commercial team wanted a proposition broad enough for clients planning regional campaigns. Zoya asked us to separate the demographic adjustment from the claim about where the evidence came from. I supported an eleven-week review and product revision costing ₹62 lakh. We traced the recruitment scope into the descriptions and outputs clients used, then restricted the geographic proposition to the eighteen-town panel. The report continued to explain that recruitment and response patterns could limit what the observed sample said about people who were not represented in it.
We did not replace the district claim with an assertion that the panel represented every resident of the eighteen towns. A narrower geographic description was necessary, but did not remove every source of uncertainty. The study could still offer evidence for carefully framed questions if the client understood its population and assumptions. A decision requiring a different geographic or population claim would need a design capable of supporting that claim rather than a more expansive description of the same records.
The trade-off was reducing the breadth of the marketed proposition while retaining the useful research within its actual boundaries. No new field coverage was claimed, and the ₹62 lakh was review and revision expense rather than a measurement of recovered revenue. The outcome was a more accurate account of what the product could support. The case reinforced a practical responsibility for my data function: definitions must remain visible when the report is sold, not only when a method specialist reads the working files. A demographic adjustment should never become a rhetorical substitute for evidence the study did not collect.