Chief Information Officer India

Where a technical completion needed another answer

Technology function through four leads: 39 staff; ₹ 8.6 crore FY 27 expense; programme separately approved
Document systems to practice-wide technology: 27 years; July 1999 entry; CIO since January 2022
Provisional annual board allowance: 24 days; Registration, fit and employer consent still to examine

Decisions & consequences

Holding twelve benchmark cases after an upgrade that passed the opening test

In 2022 I approved an application upgrade after the support team's checks showed that the practice's existing files opened and the ordinary functions were available. Sonal Drezvemi challenged that acceptance account. Her team compared 120 agreed engineering benchmark cases under predefined technical criteria. Ninety met those criteria, eighteen differed because of changed defaults and twelve had differences requiring further explanation. The file-opening result was true, but I had allowed it to answer a broader question than it supported.

The release sponsor wanted the practice to migrate together and investigate exceptions during use. Sonal objected that the professional reviewer needed to understand the changed behaviour before relying on it. I supported ₹ 92 lakh of investigation and a bounded migration decision. Technology preserved the relevant environments and comparison records; qualified practice owners investigated the results and retained acceptance authority. We did not widen the comparison limits merely because the original release timetable was becoming inconvenient.

The eighteen default-related differences were understood and accepted by the practice under its stated criteria. That left 108 supported benchmark cases and twelve outside the migration, with a maintained working route while investigation continued. The accepted cases were not described as independently certified engineering designs, and the twelve were not automatically classified as defects in the software. They were unresolved differences relevant to a professional-use decision. Any client work still required its ordinary qualified review and authorisation.

The choice prolonged support for more than one environment and delayed the clean transition the sponsor had expected. It also exposed a weakness in my own approval. I had focused on application availability because that was the part my team could demonstrate directly. The consequence belonged to the practice using the application, so its acceptance questions should have been present from the start. I would now define the technical and professional acceptance owners together before approving the deployment approach. The result I can defend is narrower than a completed transformation: we funded the investigation, supported migration where the agreed evidence existed and preserved twelve open questions instead of allowing a release date to settle them by default.

Keeping four tasks on local workstations after a remote-computing pilot

A 2023 proposal aimed to give the advisory teams a common remote-computing route. The initial discussion emphasised connection capacity and the convenience of reaching a familiar desktop from another location. Practice colleagues asked about the actual interaction: manipulating a model, reviewing a drawing and responding to a colleague while the view changed. Those tasks could not be evaluated from the advertised connection specification alone. I asked the pilot to examine representative work under the conditions in which people would use it.

We funded ₹ 42 lakh for the pilot, including user examination, configuration and support preparation. Twenty-four representative tasks were assessed against agreed interaction conditions. The platform sponsor preferred one route for everyone because it would simplify support and equipment planning. Sonal wanted the practice to retain an alternative where the working experience was inadequate. I supported a decision that followed the observed tasks, while security and software-use permissions were examined by their responsible owners rather than assumed from technical connectivity.

Twenty tasks met the agreed conditions. Four retained designated local workstations while the remote limitations were investigated. The new arrangement added ₹ 16 lakh annual support expense. We did not claim that the pilot removed the need for all local equipment or proved that the entire workforce could operate remotely. The result concerned the tested tasks and conditions. It also did not establish the quality of the engineering work produced through either route; that remained subject to the practice's normal professional review.

Keeping an exception made the operating model less uniform and left managers with some scheduling work. The alternative was to describe a technically accessible desktop as a suitable workplace for every task, pushing the consequences onto users when that was untrue. I preferred an explicit exception with support ownership. I would now include the most interaction-sensitive work earlier, alongside ordinary document tasks, so a successful introductory demonstration does not define expectations for the whole practice. The useful outcome was a bounded working option with a cost and an unfinished part, not a broad claim about productivity or a staffing reduction that the pilot had never measured.

Changing the queue without making every urgent request a permanent priority

In 2024 our shared computing queue processed work largely in submission order. The rule was easy to defend as consistent, but practice managers showed how a sequence of long jobs could delay a short calculation needed for an imminent client review. Other managers warned that allowing urgency labels would simply reward the teams most willing to escalate. Both concerns were credible. The scheduling policy was already making a business choice; keeping it implicit did not remove its consequences.

I supported a ₹ 36 lakh programme to revise scheduling and establish a practice-owned priority process. Authorised managers had to explain the dependency behind a request, while waiting work retained attention through an ageing mechanism and an escalation route. Technology implemented the agreed rules and made the resulting queue visible. It did not decide which client commitment mattered most, and a priority label did not permit a team to bypass professional review of the eventual result. The policy concerned access to computing time, not the validity of the analysis performed.

During a defined later observation period, eighty jobs were submitted for completion before the agreed cutoff. Sixty-two finished within that period and eighteen carried forward with their status retained. Those figures described that particular workload and period. We had not run a controlled comparison with the previous rule, and I did not attribute every completion to the change or offer the count as a universal turnaround promise. Practice owners still had to examine the reasons for unfinished work and the implications for their commitments.

The revised arrangement required judgement and administration that submission order had avoided. Some teams lost the comfort of assuming that an earlier submission would always be served first; others had to explain why their request warranted different treatment. We also had to resist treating every late request as an emergency. I would now agree the business policy before a shared computing service becomes heavily used, including how prolonged waiting is recognised and who can resolve competing claims. The case matters because resource allocation can appear to be a technical detail while quietly determining which part of the enterprise can keep its next promise.

Refusing to infer a measurement's unit from a plausible-looking value

A 2025 integration review examined 120 measurement rows imported from professional working material. One hundred carried explicit unit specifications suitable for the receiving process. Twenty did not. The values looked plausible, and the integration sponsor proposed a default based on the range normally seen in that field. Sonal's team objected that a plausible magnitude did not establish what the source author had meant. A successful import could conceal a change in meaning before anyone noticed a visibly unreasonable number.

We allocated ₹ 38 lakh to source review and the integration correction. The responsible source owners confirmed units for twelve of the twenty ambiguous rows. Eight remained held rather than receive an inferred label. Technology retained the confirmation basis and applied the declared representation; qualified practice colleagues remained responsible for assessing whether a measurement was appropriate for the engineering question. Confirming a unit did not establish that the underlying observation, method or design was correct.

The final account contained 112 usable rows for the defined import purpose and eight explicit exceptions. We did not call the whole source population validated, extrapolate the result to other datasets or rewrite the original material without an authorised correction. The programme changed the receiving process so that an absent specification led to a visible question. It also required colleagues to supply details that the earlier interface had allowed them to omit, adding some preparation work before a transfer could be accepted.

The sponsor's concern about friction was reasonable. Repeated manual clarification can make an integration slow and expensive. My response was to improve the source agreement and capture the required meaning earlier, rather than make the ambiguity disappear through a confident default. I had learned in document systems that faithful movement of a field was not the same as faithful communication of its meaning. I would now make units and conversion responsibility part of interface acceptance before the first production import. The outcome here was a smaller, supportable set and eight retained questions, with the cost of obtaining clarity visible in the investment account.

Discuss a leadership challenge