Chief Information Officer / India
The choices behind a more trustworthy representation
- Floor-covering retail estate
- 32 showrooms
- Fictional Nerqali Floor Stores, headquartered in Guwahati
- Internal technology remit
- 48 employees
- Twelve contractor positions are additional; five reporting leaders
- FY2026–27 delegated authority
- ₹14.8 crore
- ₹11.6 crore operating expense plus ₹3.2 crore approved change spending
Decisions & consequences
01Following Assamese and Bengali text beyond the data-entry screen
In 2022, a review of 240 customer quotes found seventy-eight with missing or broken characters in names or descriptions. Staff could see the intended text in the application, but the generated document and some printed copies did not preserve it. The problem was especially visible in Assamese and Bengali. I had accepted separate checks for data entry and file generation without requiring a reader to examine the finished customer document. That omission made a technical success look like a completed service.
The quick proposal was to ask staff to retype affected fields in Roman letters. It would have reduced the immediate rendering problem, but it would also have transferred the correction to the customer and lost information in names or descriptions. I supported a ₹42 lakh project covering document generation, appropriate font support and review of representative output, with ₹8 lakh annual support expense after implementation. Colleagues competent in the languages worked with commercial staff to distinguish broken characters from incorrect or ambiguous wording.
We tested documents through the actual delivery and printing paths used by showrooms. Technical checks could identify absent characters and failed output, while people reviewed the meaning of selected descriptions. We did not describe that arrangement as automated translation. Product descriptions and customer names needed different treatment, and a plausible string of characters could still be the wrong account of what had been agreed. Commercial head Sanjay Qervathi remained responsible for the customer wording; my team owned the production path and its technical evidence.
In a later sample of 260 documents, no missing-character defects were found, but fourteen required meaning or wording corrections before final acceptance. The samples differed and the second was not proof that every future document would render correctly. We retained a route for reporting and reviewing exceptions. The useful outcome was a more complete definition of document quality, including the work that software could not settle. I learned to follow the customer's words into the document they actually receive rather than stop at the screen where a colleague entered them.
Making two approved-looking room plans an explicit decision
A 2023 review of 119 flooring projects found twenty-two with competing room plans that different participants treated as approved. A showroom colleague might adjust a doorway detail while a survey specialist updated a room dimension. The later upload could then become the apparent current plan even when it did not include the other change. Each file looked complete on its own. The system's preference for the newest timestamp was supplying a decision that nobody had consciously made.
The application supplier proposed automatically selecting the most recently uploaded plan and keeping the older file in history. That was administratively simple, but it did not establish that the later author knew about or intended to supersede the earlier agreement. I proposed explicit revision branches and a reconciliation step owned by the responsible project specialist. The work required ₹76 lakh capital and ₹19 lakh implementation expense. Finance determined the accounting treatment; the proposal distinguished those amounts rather than presenting them as one recurring technology cost.
The new workflow retained the relevant versions, showed which changes were in dispute and required an identified reviewer to establish the issued plan. The application did not merge room geometry or certify measurements on its own. Where a new visit or customer clarification was required, that work remained visible. Some colleagues worried that the process would create more exceptions and delay installation. I accepted the need for a response standard, but argued that hiding a disagreement in a last-upload rule did not remove the underlying risk or the time needed to resolve it.
During a six-month review ending September 2024, nine of 150 projects required explicit reconciliation before issue. Those nine were recorded as work caught and reviewed, not as nine defective installations or proof of a statistically comparable reduction from the earlier sample. The main result was that an unresolved difference no longer silently became an instruction. I remained accountable for the system behaviour, while qualified colleagues owned the room measurements and the final technical decision. The case changed my view of simplicity: fewer visible exceptions can mean less useful information, not necessarily a better process.
Reviewing what remote consultations left behind
Our remote room consultations accumulated 1,830 recordings by the January 2025 review. Within that collection, 642 related to projects closed more than twelve months earlier. The recordings could include views of customers' homes and conversations that were not required in the final order record. Cheap storage had made continued retention easy. It had not supplied an approved reason for retaining every recording or leaving access as broad as it had been during the consultation.
I asked for a purpose and preservation review with the relevant legal, privacy and commercial advisers. We allocated ₹31 lakh to classification, copy discovery, access changes and implementation of the approved actions. One suggestion was to delete everything older than a year. Another was to keep the entire collection indefinitely in case it became useful. I supported neither as a blanket rule. The time since project closure identified a review population; it was not a claim about a universal statutory retention deadline or permission to disregard an identified preservation obligation.
For the 642 older closed-project recordings, the authorised review resulted in 508 approved deletions, ninety-four restricted records with documented continuing purposes and forty under specific preservation holds. The categories totalled the original population and were kept distinct in the report. The team addressed relevant copies and permissions through the approved process. It did not delete the underlying order or financial records merely because a consultation recording was removed. The other 1,188 recordings remained subject to their own review schedule and were not implied to have been cleared by this exercise.
The approved actions for that population were completed by June 2025. I reported the classification and execution evidence without calling the project a legal-compliance certificate or a financial saving. Continuing purposes and holds still needed periodic review by their proper owners. The experience made me more careful about services that create rich information as a by-product. Before expanding another consultation feature, I now ask what it captures, which part is needed after the conversation and how the company will recognise when that need has ended.
Narrowing a room visualiser after customers read too much into it
I supported a ₹1.4 crore pilot of a room visualiser in 2026. It helped customers discuss how a floor covering might relate to furniture and room layout. The demonstration was persuasive, but the underlying service could not establish an exact physical shade under the lighting and viewing conditions in each customer's home. In a review of 176 pilot viewers, forty-three understood the image as an exact shade preview. That interpretation was more consequential than the internal description of the feature as illustrative.
The first proposed response was a clearer explanation alongside the image. Commercial colleagues reasonably wanted to preserve a tool that made consultation more engaging, and I wanted the investment to have a useful result. We changed the explanation and the demonstration sequence. In a later group of 180 viewers, twenty-six still treated the image as an exact shade preview. These were separate groups, not a controlled experiment, but the remaining misunderstanding was enough to challenge the proposition. We could not rely on a sentence to reverse the confidence created by the image.
I recommended retaining layout exploration while removing the unsupported shade-preview claim and the interface treatment that encouraged it. Product specialists and commercial colleagues established where physical samples and their advice remained necessary. We spent a further ₹18 lakh revising the service and staff guidance. That amount was additional to the pilot, and neither sum was presented as recovered merely because the narrower tool remained in use. CEO Nusrat Delvani supported the reduced scope rather than treating money already spent as a reason to retain a stronger promise.
At the October 2026 review, the restricted layout service was available, but we had not established a separate commercial return or completed a new interpretation study. I kept those limits in the account. The decision was not that digital visualisation has no value. It was that this service needed to be described and operated within what it could support. I had helped make the original demonstration attractive; I also had to help remove a confidence it had not earned.