Chief Information Officer / India
The convenient system rule that changed the meaning
- Information colleagues
- 22
- Three managers; ₹6.4 crore annual operating expense
- Institutional customers
- 23
- Independent biomedical-equipment servicing contracts
- Information experience
- 27 years
- Service applications to enterprise information leadership
Decisions & consequences
01Removing an automatic approval hidden inside visit closure
In 2022, a review of sixty-four completed service visits found nineteen records without the separate engineering acceptance required by the relevant workflow. The application had mapped administrative visit closure into a field that reporting treated as technical acceptance. The visits had been marked complete by the service administration team. That did not establish that the qualified technical owner had made the separate decision. The review concerned a defined set of records, not a measured failure rate across all customer equipment.
I had approved the interface mapping after a design discussion that used the word complete too casually. The service administration manager wanted to keep the single status because it reduced follow-up calls and made the daily list easier to clear. Joseph Melqashi challenged the meaning of the report. I accepted his objection and proposed removing the automatic mapping, retaining separate states and permissions, and presenting unresolved acceptance explicitly. Technical colleagues would review the nineteen records through their own process; my application team would not infer approval from the absence of an objection.
The eight-week correction cost ₹24 lakh, covering application changes, record review support, testing and staff training. We retained the previous entries so that the reason for correction remained visible. During the transition, service administrators had to check a separate list and explain outstanding decisions to account teams. That was additional work, and some customer reporting took longer. I included the temporary burden in the proposal rather than promise that the new screen would immediately simplify every task.
The revised workflow separated administrative completion from engineering acceptance and prevented the old mapping from supplying the latter. The technical disposition of individual records remained with qualified owners and the applicable customer arrangements. We did not describe the change as a device-safety certification or claim a clinical benefit. The lesson for me was that an interface can transfer a field correctly while transferring the wrong meaning. Approval of a mapping requires an explicit account of the decision it represents, especially when two teams use the same everyday word for different responsibilities.
Restoring fourteen equipment histories from seven merged identities
During a 2024 migration review, the technical team found seven serial strings that appeared on equipment from two different manufacturers. Fourteen assets had been reduced to seven application records because the import treated the serial string alone as a universal identifier. The merged rows contained events that did not all belong to the same device. A successful import count had therefore concealed an identity problem. We could not repair it merely by changing the display name or duplicating the merged history into two new records.
The implementation supplier proposed adding a manufacturer prefix and copying the existing event list to both resulting records, then asking service teams to correct exceptions later. That would have made the migration look complete while knowingly attaching uncertain events to equipment. I rejected the shortcut. We agreed an identifying structure with technical owners, created separate asset records and required verification of the event-to-asset links. Where the source evidence did not establish a link, the event remained flagged for investigation rather than silently assigned. Customer representatives were involved where their asset records were needed.
The programme took nine weeks and cost ₹27 lakh for reconciliation support, interface changes, technical review time and migration testing. It delayed the planned retirement of the old application and required a controlled read-only reference period. The supplier disputed responsibility for some of the extra work because our original specification had not stated the cross-manufacturer condition. I acknowledged that omission and took the proposed allocation of cost through the agreed commercial process instead of treating the entire repair as a vendor failure.
The fourteen assets were restored as distinct records with verified links where evidence supported them; unresolved historical links stayed visible. We also tested the same identifying rules at search, import and interface boundaries. That did not prove every record in the wider estate was correct. It addressed a specific failure and changed the checks for subsequent migration work. I now ask what makes an identifier unique within its actual context before allowing a convenient field to determine the identity of a serviced asset.
Keeping both sides of a field-edit conflict available for review
In 2025 we tested forty-eight offline synchronisation scenarios before extending a field-service application. Twelve showed an older local edit replacing a newer server entry when the device reconnected. The software used arrival order to choose the surviving version. A field engineer could therefore upload legitimate work from an earlier visit after a supervisor had recorded a later correction, leaving the screen with a misleading sequence. The exercise identified a design defect; it did not establish twelve actual equipment incidents or a clinical outcome.
The delivery team preferred to retain the existing rule and train engineers to reconnect before editing. That would have made reliable connectivity an unstated condition of the service, despite locations where it could not be assumed. I had supported the original release plan and withdrew it. We preserved both versions when a conflict was detected, displayed their origin and timing, and routed the substantive choice to an authorised technical owner. The resolution record had to show who decided and the reason. My information team could operate the mechanism without deciding which engineering finding should prevail.
We spent ₹19 lakh over six weeks on the change, testing and support preparation. The revised workflow created a visible exception queue. It required time from technical supervisors and could delay the point at which a service record became settled. I asked Joseph Melqashi to estimate that workload and included it in the release decision. We did not promise that the application would remove every disagreement or automatically determine the correct history from timestamps alone.
The same forty-eight exercises then retained conflicting versions and followed the agreed resolution route. Wider use still required monitoring and customer-specific arrangements where relevant. I reported the test scope, the remaining support obligation and the conditions for extension separately. The important correction was to stop treating the last arriving message as the last valid decision. A service application should make a difficult disagreement available to the responsible people, rather than conceal it through a technically simple overwrite rule.