Chief Information Officer / India
Keep the unfinished task in the report
- Internal-technology colleagues
- 33
- Four managers; ₹14.6 crore annual operating expenditure
- Company employees
- 520
- March 2026; ₹178 crore FY2025–26 revenue
- Internal-systems experience
- 25 years
- Learning-centre applications to CIO responsibility
Decisions & consequences
01Withdrawing four answers that search could not make current
In 2023 we reviewed 120 practical queries employees used to find internal guidance. The search service was fast and its index contained the intended document collections. Seventeen queries nevertheless led first to guidance that had been superseded. Some older pages had accumulated links and familiar wording, so they appeared more useful to the ranking system than the current material. A quick answer was creating an impression of authority that the source no longer deserved.
I had supported the broad indexing approach and initially expected a relevance adjustment to be sufficient. The review showed that part of the problem belonged to the source itself. We worked with departmental owners to identify the current answer, the status of older material and the route for questions that remained undecided. Thirteen of the seventeen first results were replaced with owner-approved guidance. Four were withdrawn from the current-answer route while the responsible department resolved its position. The older records remained available in an appropriately labelled archive where required; we did not erase history to make the search report look cleaner.
The correction programme cost ₹24 lakh and took eight weeks. It combined source ownership, status fields, result presentation and tests using the reviewed queries. We reported the defined set corrected and the four unresolved answers still awaiting business decisions. We did not claim that all possible employee searches were now accurate or multiply a small task exercise into an annual productivity saving. Search behaviour can change as new material and new questions appear, so ongoing ownership mattered more than a one-time improvement score.
The trade-off was that some users received an explicit unresolved answer where they had previously received a confident-looking document. I considered that a better service, provided we named the owner and next step. The case changed my view of completeness. Adding every available page to an index can make the system technically comprehensive while making its answer less trustworthy. We now ask what status the reader needs to see and who is accountable for keeping that status meaningful, before describing a larger collection as a better knowledge service.
Reopening requests that our closure rule had hidden
Our employee-support process automatically closed certain requests after a defined period without a reply to a support message. The rule had been introduced to prevent old conversations remaining open indefinitely. In a 2024 review of 210 requests closed through that route, we found forty-four where the employee's original task was still unresolved. Some people were away; others had found a temporary workaround or had not understood that the message required confirmation. The closed status did not reliably describe restored service.
I asked Grace Dolqeni to reopen the forty-four requests and give each an accountable service owner. We agreed how to contact the employee through an appropriate available channel and distinguish confirmed recovery, an accepted workaround and a problem still preventing work. The remaining 166 reviewed requests did not belong in the same unresolved group on the evidence examined, but we avoided describing the sample as proof that every automatically closed request elsewhere was sound. We also preserved the original closure history so the review could explain what had happened.
Within ten working days, thirty-eight of the reopened requests were resolved and six remained open with owners and agreed next actions. We spent ₹7 lakh on workflow changes and staff time for the correction. The revised process retained a route for administratively closing inactive conversations, but separated that action from a claim that the underlying task had recovered. The dashboard initially looked worse because uncertain and reopened work remained visible. I accepted that consequence instead of asking the team to regain the old closure rate through another shortcut.
The decision was partly about incentives I had helped establish. We had celebrated a reduction in the old queue without examining enough of the work behind it. A service measure should help us find the employee who still needs assistance, not make that person disappear from the report. We now review a small, defined set of closed requests with task context and use the findings to improve the process. The exercise did not establish a universal resolution rate or a financial saving. It corrected specific work and made the difference between process closure and useful recovery easier to see.
Keeping four identity matches outside a completed integration
When a small delivery office joined our operating organisation in 2025, the integration plan covered 120 employee records. The technical team could import names, email addresses and current roles into the shared directory. A closer comparison found thirty-one ambiguous matches with existing or historical records. Some people had changed names or addresses; others shared similar names. A proposed automatic merge would have made the new directory look complete but could have confused the identity behind earlier approvals and application activity.
I held those matches outside the automated merge and brought people, application and security owners into a documented reconciliation. We required evidence appropriate to each record and kept current display details separate from the historical identifier used by the originating system. The purpose was not to preserve every old account as active. Access changes followed their own authorised process. The work concerned who a record represented and how a reader could understand an earlier decision after the organisation's current details had changed.
Twenty-seven of the ambiguous matches were resolved within the planned review period. Four remained separate while additional evidence was obtained. We accepted a six-day extension to that part of the integration and spent ₹16 lakh on review and application changes. The other eighty-nine employee records had passed the defined matching checks; they were not evidence that the unresolved four could safely be inferred. We reported the remaining boundary to the integration sponsor and provided appropriate interim service arrangements without inventing an identity conclusion.
The case required a different definition of completion from the one in the original project schedule. A current employee could be supported while a historical match remained under review, provided the services and permissions were handled deliberately. Conversely, a populated directory could still conceal an unresolved attribution problem. I changed the acceptance discussion to include the people who owned the records and the consumers of historical approvals. The result was a more defensible account of the integration, not a claim that name matching had become infallible or that a data import itself established authority for every downstream action.