Chief Data Officer / India
Correcting the meaning without inventing missing evidence
- Data employees
- 34
- Three heads; company employment totals 380
- Annual data operating expense
- ₹5.9 crore
- Separate ₹1.4 crore evidence programme
- Career since June 1997
- 29 years
- Agricultural research records through bioenergy data leadership
Decisions & consequences
01Correcting the residue total in a dashboard I had approved
In 2018 I approved a supply dashboard whose reviewed forty weighbridge records totalled 420 tonnes under the heading received residue. The source field contained gross loaded-vehicle weights. Corresponding tare weights totalled 120 tonnes, leaving 300 tonnes of net material. The import had reproduced the source values accurately, and that successful transfer had been central to my approval. Tara Vaqzelri’s review challenged the meaning of the displayed measure rather than the accuracy of the copied values.
The programme team wanted to change the label quietly at the next release because the underlying records remained available. I did not think that addressed the current use of the dashboard in procurement planning. I withdrew the affected view and explained the correction to the responsible business owners. The difference was not 120 tonnes of missing physical residue. It was the difference between the weight of loaded vehicles and the material they carried. Financial or supplier consequences required their own examination instead of being inferred from the revised display.
Management approved ₹29 lakh and seven weeks to rebuild the measure and its acceptance checks. The revised account preserved gross, tare and net values and showed how the total was obtained. All forty records were reconciled under the reviewed definition. Qualified owners confirmed the scope of the planning use before release. We did not describe the corrected quantity as dry material, energy content or an accepted procurement value, because those would require additional evidence and decisions. The correction established a more faithful weight account, not every conclusion someone might wish to draw from it.
I told the team that my own acceptance standard had been too narrow. Matching a source field did not establish that the field meant what our interface said it meant. That changed the questions I asked at subsequent releases: what physical quantity is being represented, what transformation occurred and which owner can confirm the resulting meaning? The case remains useful because the error did not require a complex model. A confident label was enough to turn correctly transferred data into an unsupported operating assumption.
Keeping four lot histories unresolved after reconstructing the other fifty-six
A 2023 review followed sixty outbound residue lots through the recorded supply chain. Forty-two had adequately supported histories under the agreed review definition, while eighteen contained ambiguous links. Some incoming lots had been split, and some outbound lots combined material from several sources. An inherited single-parent field could not represent every relationship without losing part of that history. Filling it consistently would have made the record easier to display while making the physical account less accurate.
The operations sponsor wanted a complete lineage column for a forthcoming customer discussion. I proposed keeping the eighteen ambiguous histories visibly unresolved while the source records were examined. Quality and operating owners helped define the events and quantity relationships needed for the intended enquiry. We did not infer a unique origin from a convenient storage label or assume that a shared date proved a link. The work required an account of what had actually been recorded, including where the record was insufficient.
Management authorised ₹47 lakh and eleven weeks for reconstruction and a revised representation of split and merged lots. The bounded review supported fifty-six histories and left four unresolved. The four did not become traceable through a default value, and the improved proportion was not described as universal traceability across the entire enterprise. The revised records supported the defined quality enquiry. They did not independently establish material suitability, certification, sustainability or emissions performance, all of which could require different evidence and qualified assessment.
The outcome was a more useful chain of relationships with an honest boundary. A customer-facing account could explain which reviewed dispatches had supported histories and which still needed work. I wanted the unresolved four to remain visible after the programme closed, because otherwise the pressure for a complete column would simply return at the next presentation. The strategic value of the correction was not the appearance of a perfectly connected database. It was the ability to follow the material through the transformations the business actually performed without inventing a simpler past.
Separating moisture bases before comparing laboratory returns
In 2025 a proposed comparison brought together twenty-four laboratory moisture results. Sixteen were reported on a wet basis and eight on a dry basis. The draft table placed their percentages in one column without preserving that distinction. A higher value could therefore reflect a different denominator rather than wetter material. The source results were not necessarily wrong; the proposed comparison was missing a condition needed to interpret them.
Tara Vaqzelri illustrated the issue with a constructed sample containing 20 units of water and 80 of dry matter. Water was 20 percent of the 100-unit total, but 25 percent of the 80-unit dry component. Those two percentages described the same sample. I withheld the common ranking and asked laboratory owners to review basis, sampling and method context before the values were treated as comparable. The team initially wanted a formula-only conversion. I agreed that conversion could address the denominator, but it could not settle whether the samples and methods supported the same comparison.
Management approved ₹38 lakh and eight weeks for the evidence review and revised presentation. Twenty-two records supported the defined comparison after qualified review; two remained outside it because their supporting context was insufficient. Original results and bases remained available alongside any reviewed derived values. We did not overwrite the laboratory account or prescribe a material-acceptance threshold. Procurement, quality and customer-operating decisions retained their owners, and no claim about fuel performance or safe use was inferred from the percentage alone.
The result was a narrower comparison with a clearer meaning. Two records remained visible as unresolved instead of being forced into a ranking merely because every row contained a number. I wanted users to understand why a mathematical conversion was necessary but not sufficient. Data leadership in this setting means keeping the sample and the denominator attached to the figure as it moves into a business decision. A clean percentage column should not erase the conditions that determine whether its rows can responsibly be compared at all.