Leadership review
Kavita Zevqumi
Chief Information Officer · India
Highlights
47
Technology colleagues through four functional heads
₹19 crore FY 2026–27 operating expense; ₹7 crore capital separately
31 years
Programming through service leadership
Five continuous appointments beginning July 1995
620,000
Member records administered, not assets managed
186 employer-trust clients; no client investment assets held
A letter from Kavita Zevqumi
I want the system to preserve the explanation a person may need years after the original transaction. Retirement administration brings together employment history, scheme rules, authorised decisions and records that continue to matter after active service ends. A current screen can look orderly while losing the version or relationship needed to explain an older calculation. My job is to make that continuity workable without pretending that software owns the client's judgement.
The first lesson I carried from programming into leadership was that a current value can be an incomplete explanation. At Temzuri Data Services in Bhilai, I worked on administrative data used by business-service teams. A field could be correct today while leaving a colleague unable to explain what had changed since the last statement. I learned to ask which event the record described and who would need to understand it later. That habit became more consequential when I moved into retirement-benefit administration, where the useful life of a record can extend well beyond the employment relationship that created it.
Kavita Zevqumi
Chief Information Officer
Strategic priorities
01
02
03
04
Performance
Priority outcome 01
A 2022 review examined 1,260 historical benefit-calculation records for a defined group of client schemes. Eighty-four lacked a sufficiently linked reference to the rule version used at the time. The current rule library was available, and the application team proposed attaching the latest version to complete the record. Benefit operations challenged that approach because a current rule could not explain a calculation made under an earlier effective period. The gap concerned evidence; it did not establish that eighty-four members had received the wrong benefit. I supported Nazia Tezqavri's request to retain the records for authorised review. We spent ₹48 lakh on controlled evidence reconstruction, application changes and review support. The project team wanted to close the work with a technical update, while client-service colleagues were concerned about the effort required from trust representatives. I explained that replacing a missing historical reference with a convenient current one would make the system appear complete by changing the question it answered. The teams located supporting rule versions and decision records for sixty-seven of the eighty-four cases. Seventeen still required client clarification at the reporting date. Benefit operations and the relevant client owners assessed the evidence and any resulting action; technology did not approve a revised entitlement or payment. We changed the calculation record so that the applicable version, effective period and authorised basis travelled with the result, with controlled amendment rather than silent replacement. The consequence was additional expense and an unfinished group remaining visible. I did not describe the sixty-seven reconstructed records as corrected benefits, because the work established an explanation rather than a uniform financial error. The seventeen unresolved cases were not assumed to be wrong simply because their evidence was incomplete. I would specify the historical reference as part of the original record design, before a later review has to rebuild it. A retirement-administration system needs to preserve the basis of a decision, not merely keep a plausible current rule beside an old number.
Priority outcome 02
A 2023 support-workspace rehearsal used synthetic member records across twelve fictional client environments. The interface separated clients, but testing showed that three of forty-two export actions could include records outside the support user's authorised client scope. No live member data was used in the rehearsal, and the finding was not evidence of a historical disclosure. It did establish that a visible client selector was insufficient to enforce the boundary on every action. The support manager wanted the workspace released with a warning because it would shorten the time needed to answer client enquiries. Security colleagues recommended retaining the existing restricted tools until the export path was corrected and retested. I supported the delay. We deferred release by five weeks and spent ₹36 lakh on correction, independent testing and temporary service support. The cost included keeping a less convenient process in use while the proposed improvement remained incomplete. Application engineers moved the authority check into the server-side export operation and examined the related request paths. Business owners confirmed the permitted support scope rather than leaving the developers to infer it from a screen design. The retest covered all forty-two actions and the defined attempts to cross client boundaries. Those checks passed before the revised release was approved through the company's process. Security retained responsibility for the broader assessment and continuing monitoring. The result was a supported release decision for the tested scope, not proof that every possible access weakness had been eliminated. I kept the synthetic nature of the exercise and its limits in the management account. We had found a route that should not have existed and paid to remove it before using the workspace with live records. My earlier confidence in the client-separated interface had been too visual. I now ask what authority each consequential action checks, including an export that can travel well beyond the screen on which it was requested.
Priority outcome 03
In 2024 I approved a statement programme that initially treated download availability as the principal improvement. Members could obtain a file without waiting for paper, and the delivery report made that sound like a completed access problem. A rehearsal with twenty-eight participants using assistive tools challenged the conclusion. Eleven could not complete the defined sequence of finding the relevant period, reading the calculation explanation and locating the enquiry route on the first version. The difficulty was not resolved by proving that a file had downloaded. The delivery team preferred to publish the scheduled version and improve it later. Member-service colleagues argued that the statement would become the starting point for consequential questions, including questions from people no longer able to use a former employer's internal support. I accepted responsibility for the earlier narrow criterion and approved ₹27 lakh for document-structure changes, testing and assisted-service preparation. The release moved by four weeks. The team revised reading order, headings and navigation, with benefit specialists checking that the explanation retained its meaning. A second rehearsal using the same task definitions left three participants needing assistance. We kept that result in the release account and maintained an assisted route instead of presenting the revision as independently usable by everyone. The exercise did not establish performance across every device, disability or member circumstance, and we did not treat the participant group as a statistical estimate for the entire population. The decision produced a better-supported statement and an ongoing service cost. It did not make the assisted route a failure to be hidden once a digital file existed. I learned to ask what a person must be able to do with the statement before deciding which completion measure belongs in the project report. Technology could improve the document and its delivery; the organisation still needed to explain the benefit through authorised people when the member required help. Availability was one part of access, not the whole result.
Priority outcome 04
A 2025 document programme examined 2,400 scanned nomination forms in a bounded client-authorised review. The proposed extraction tool could identify text and populate fields quickly. Its demonstration encouraged the idea that a readable output was ready to become an authoritative record. Benefit operations found 312 forms with ambiguity requiring review, including overwritten entries, incomplete relationships and marks whose meaning could not be established from the extracted text alone. Those were review cases, not evidence that every nomination was invalid. The automation sponsor wanted to accept high-confidence fields automatically and send only low-confidence text for checking. Nazia Tezqavri argued that a clear word could still belong to an ambiguous instruction. I agreed to use extraction as preparation for the authorised review, with the source image available beside the proposed fields. The company spent ₹42 lakh on review support and the controlled workflow rather than record the full proposed labour reduction as achieved. Within the 312-case group, 226 were resolved through the approved client process and eighty-six still awaited clarification at the reporting point. The remaining 2,088 forms followed their normal checks; their exclusion from the ambiguity group did not waive those checks or create a technology certification of entitlement. Client trustees and authorised administrators retained the relevant decisions, with qualified advice where necessary. The application preserved the source, reviewer action and decision basis rather than silently replacing the image with machine-produced text. The result was a narrower automation claim and continuing manual work. I did not call the eighty-six pending cases processing failures merely because the technology could already read parts of them. The unresolved issue concerned the instruction and the evidence required to act on it. I would involve benefit operations earlier in defining what a completed record means. Extraction can reduce clerical effort, but a confidence score about characters cannot settle the authority attached to a person's retirement-benefit instruction.
Priority outcome 05
A 2025 software renewal offered a lower fee if the agreement counted only members currently employed by client organisations. Our defined record population contained 410,000 current-employment records and 210,000 former-employee or retired-member records. The second group was not uniformly active every day, but it still required authorised enquiries, statements and continuing administration. A price based on the first group alone could narrow the permitted service in a way that the commercial comparison initially concealed. Procurement preferred the ₹2.8 crore annual proposal because it aligned with the supplier's active-user definition. Benefit operations explained the tasks still required for the other records. I asked the supplier, legal and commercial teams to make the permitted actions explicit before comparing prices. We chose a ₹3.5 crore annual agreement covering the defined service capabilities across both populations, accepting a ₹70 lakh annual premium. The cost was a deliberate choice, not a saving against a hypothetical future dispute. The agreement distinguished current-employment activity from the lower-frequency work required for other members, rather than pretending that all 620,000 records represented identical daily users. Technology checked that the system could support the agreed tasks and access routes. Procurement and legal colleagues retained responsibility for the contract, and finance assessed affordability. We also retained the population definitions in the renewal review so a later change in employment status would not automatically remove a necessary capability. The immediate outcome was a broader supported service at a higher annual cost. I did not claim that the premium improved every member's experience or that every former employee would use the system during the year. The decision concerned the obligation the company needed the agreement to permit. I would bring benefit operations into the first pricing discussion next time, before an attractive unit definition becomes the assumed description of the business. A person can leave employment while the administrator's responsibility to maintain a usable record continues.
Stakeholder voices
Kavita kept the historical evidence gap separate from a conclusion about the member's benefit. She funded the work needed for an authorised explanation and retained the unresolved cases in the board account.
I work with Kavita on benefit operations. She understands why a readable field or a current rule does not settle every instruction, and she makes space for the client decision instead of asking software to borrow that authority.
Outlook
I want a record to preserve the basis of a calculation, including the unresolved questions, rather than acquire a convincing explanation from whatever rule happens to be current.
Read the essayI had allowed delivery of a file to stand for a member's ability to use it. A small rehearsal changed both the release decision and the service we kept around the document.
Read the essayA separate client workspace looked convincing until we tested its exports. The release decision changed when we examined what the action could do, rather than what the interface appeared to permit.
Read the essayAn invitation
I am considering a first outside board contribution where long-lived records and continuing customer obligations matter. I have no statutory appointment and have not yet registered in the IICA databank. Preparation remains unfinished. Twenty-five days is a provisional capacity allowance, subject to employer consent, relationships and the demands of a specific role.
office@kavitazevqumi.example