Chief Data Officer / India
A cleaner total must still describe what happened
- Data colleagues
- 39
- Three heads; ₹10.8 crore annual operating expense
- Company revenue
- ₹96 crore
- FY 2025–26; 230 employees in March 2026
- Publishing-data experience
- 30 years
- Distribution records through software data leadership
Decisions & consequences
01Withdrawing a dispatch total presented as reader sales
In 2022 a product summary described 16,000 copies dispatched into a retail channel as reader sales. The figure came from a valid movement record, but the label answered a different question. I had accepted the summary during product review without asking whether the underlying evidence followed copies through the channel. The issue became visible when a customer specialist compared the report with opening stock, returns and the stock still held at the end of the period.
We withdrew the sales label and rebuilt the reviewed account with the customer. The defined period began with 1,000 copies in channel stock and received 16,000 additional copies. The documented outcomes were 9,700 reader sales, 2,400 returns to the distributor, 600 promotional copies and 4,300 copies in closing stock. Those quantities reconciled to 17,000 available copies. Reader-sale records supported the 9,700 figure; we did not infer it solely as whatever remained after subtracting the other categories. The reconciliation was useful because both the movement account and the sale evidence could be examined together.
The correction and related product work cost ₹26 lakh over ten weeks. We kept dispatch measures available for the operational questions they could answer and revised the naming and source explanation for the reader-sales view. Channels without the necessary evidence were shown as incomplete for that purpose. We did not fill missing reader sales from dispatches or assume that every unreturned copy had been bought. The review did not certify the customer's revenue recognition, royalty calculation or financial statements. It corrected a defined software interpretation and the affected report.
The lesson was not that a dispatch count is a poor measure. It is useful when the question concerns dispatches. The error was allowing a convenient available number to stand in for a commercially more attractive answer. I took responsibility for the approval and for the expenditure needed to repair it. In later product reviews I ask the team to name the event behind a measure and to show the evidence required for the question the label invites a reader to answer.
Keeping separate products inside a shared catalogue title
A 2024 catalogue review examined 420 product groups assembled from customer records. Fifty-four were flagged because their descriptions and identifiers did not support the grouping confidently. A proposed cleanup would have collapsed the suspect entries by title and author, removing what looked like duplication. That would have made the catalogue shorter while risking the loss of distinctions important to ordering, returns and customer choice.
I asked our data team and the customer's catalogue specialists to review the 54 groups against their source records. Nineteen were confirmed duplicate descriptions of the same product. Thirty-five contained separate products that needed to remain distinct, including different bindings or editions. The title relationship could still be represented for discovery, but it could not replace the identity used for a particular saleable item. We retained the source identifiers and the review basis. Where the source did not settle an association, the product could show an unresolved relationship rather than manufacture a confident merge.
The review and model correction cost ₹21 lakh over eight weeks. We changed the grouping workflow so that a proposed consolidation showed the fields it would discard and required an appropriate review before affecting commercial records. We also preserved links between related items where the evidence supported them. That allowed a user to discover another format without treating it as the same orderable product. The 54 suspect groups were a defined review set, not evidence that the rest of every customer's catalogue had been comprehensively verified.
I had supported an earlier plan that treated a smaller number of records as a useful sign of progress. The case changed that discussion. Fewer rows can represent less repetition, or they can represent lost meaning. The team needed to explain which result it was producing. We did not claim that product identifiers settled copyright, commercial entitlement or the accuracy of every description. Those matters retained their own owners. Our responsibility was to preserve the distinctions necessary for the software's stated purpose and make any consolidation decision inspectable.
Showing what was known when the report was issued
In 2025 a review of a 5,000-line reporting extract found 312 records that had arrived after the original report for the period was issued. The product's refresh process would have replaced the earlier view silently. A user returning to examine a past decision could therefore see a different account without knowing that some of the underlying evidence had been unavailable at the time. The revised view might be useful, but it was not the report the decision-maker had originally read.
I asked the team to preserve the issued snapshot and create an explicit revision path. Customer specialists reviewed the 312 late records. Two hundred and sixty-six were accepted for the revised operational view; 46 remained unresolved. The revised report showed its preparation date, the accepted additions and the unresolved status. We retained the original snapshot for comparison instead of overwriting it or pretending that its contents had included the late evidence. Source event dates and receipt dates remained separate so a user could understand both the period being described and the arrival of the record.
The change cost ₹19 lakh and took eight weeks. We worked through the defined extract and reporting path rather than claiming a universal reconstruction of every historical report. The customer retained responsibility for any financial reporting or commercial adjustment; a dated operational revision did not itself authorise one. We also avoided treating the 46 unresolved records as zero or quietly assigning them to the period that produced the preferred result. Their status stayed visible pending the appropriate clarification.
The design had originally favoured a single current answer because that seemed easier for users. I had accepted that simplification. The review showed that the user sometimes needed two answers with different meanings: what was available then, and what the accepted evidence supports now. We changed the product explanation and the underlying retention of reports accordingly. The correction did not make the earlier decision right or wrong. It made a fairer examination possible by preserving the evidence actually available and the later changes to the account.