A deposit migration can reconcile to the last unit of principal and still change what happens to the account at maturity. The balances answer an essential question about value. They do not, by themselves, establish that the customer's instructions have retained their meaning. I regard that distinction as a central responsibility in banking technology leadership. A platform replacement should not quietly turn a technical mapping choice into a different treatment of a banking relationship.
The risk is easy to understate because a matching total provides a persuasive piece of evidence. It is concrete, familiar and straightforward to explain to a steering committee. A difference in renewal treatment can be less visible until someone examines the behaviour of individual categories. Neither check should displace the other. My argument is that the release decision needs both, with an explicit account of who understands the business meaning and who has verified the conversion's technical operation.
Arithmetic agreement leaves a second question open
Before a migration, I ask which stored fields affect a later action. The list may extend beyond the fields used to calculate today's balance. An instruction can determine what happens when a term expires, which account receives a movement or whether a customer has requested a particular treatment. The relevant service specialists should identify those meanings and the evidence needed to preserve them. Technology can then explain how the old representation becomes the new one without deciding the business interpretation on its own.
At Lanshu, our rehearsal reconciled principal across 250,000 term-deposit records. A separate review found incorrect renewal mapping in 6,200 records. The aggregate success had not been fabricated; it was simply insufficient. That finding was most useful when presented as a missing dimension of the acceptance evidence, rather than as a contest between engineers who had done their checks and operations colleagues who had discovered something else. Both groups needed a release definition broad enough to describe the service they were changing.
The experience also changed how I think about sampling. A large sample selected without regard to instruction categories can miss a smaller group whose meaning differs materially. Business specialists can help identify which categories require deliberate attention. This is not a universal statistical prescription; the design should reflect the actual population and purpose of the test. The point is to avoid treating volume alone as a substitute for understanding which differences could alter the customer's subsequent experience.
I want the conversion record to retain the reason for each material mapping decision. Future maintainers may otherwise see only that a field was assigned a particular value. They need to know which business interpretation supported it and who approved that interpretation. The record should be usable without recreating every conversation from the programme. Good documentation here is part of preserving the bank's knowledge, especially when the old system's experienced staff will move on after the transition.
Give business interpretation a place in the release decision
A service owner needs an effective opportunity to challenge the conversion before the launch date becomes politically immovable. Inviting a specialist to a meeting is not sufficient if the programme has already treated every unresolved question as a minor cleanup item. I ask the release paper to state which evidence remains incomplete, what that incompleteness could change and who can accept or reject the proposed treatment. That creates a decision a senior sponsor can understand rather than an unexplained technical objection.
Our postponement lasted six weeks and added CNY1.9 million of parallel-operation cost. Those consequences belonged beside the finding. The additional expense did not make the original mapping acceptable, but it mattered to the choice of remedy and timing. I would not claim that every delay is a sign of strong governance. A delay should have a specific purpose, a bounded plan and evidence that addresses the condition that made proceeding unacceptable. Otherwise it can become an expensive way of postponing the same uncertainty.
The CIO should also resist becoming the final interpreter of every disputed instruction. Engineers may understand the data representation better than anyone else in the room, yet the decision about the customer's banking treatment belongs with the appropriate business authority. Legal, compliance and risk specialists may need to examine particular questions. Technology's contribution is to make the options and consequences clear, implement the approved treatment and retain evidence of what was done. That division of responsibility remains valuable even when it makes the programme slower.
Disagreement can reveal a weakness in the old process as well as the new conversion. An instruction may have been recorded inconsistently for years. The migration does not acquire permission to resolve that history through whichever default is easiest to implement. Management may need a separate remediation decision. I would distinguish records ready for conversion from those requiring further authority, and ensure that the held group has an owner. An exception process is meaningful only if the bank can see the work and follow it to a decision.
Keep rehearsal findings separate from live exceptions
Correcting the 6,200 rehearsal records did not mean the eventual production transition would contain no exceptions. Our live process accepted 249,983 records through the main route and held seventeen separate cases for resolution over two days. The two populations arose at different stages and should not be combined into a narrative suggesting either that the rehearsal errors remained unresolved or that the production process was flawless. The board needs a chronology of evidence, not a single success percentage assembled from unlike events.
I favour a post-transition account that explains what the earlier testing caught, what happened during release and what remains under observation. Some consequences appear only when a later event occurs. Where that is relevant, the review should state how the bank will examine the subsequent behaviour rather than declaring final success immediately after data movement. The duration and scope of that observation should follow the actual service, with appropriate ownership and practical limits, instead of becoming an indefinite promise that nothing can go wrong.
For future programmes, I would bring instruction-level review into the first rehearsal design. That changes the work estimate and the participation required from business specialists. It may expose a less convenient timetable, but it does so while choices remain open. A schedule built around incomplete acceptance criteria does not become realistic merely because senior management has approved its date. The CIO should help the sponsor understand what evidence the schedule must produce, not just how quickly the technical transfer can run.
A bank changes platforms to improve its ability to serve, operate or adapt. Those purposes lose credibility if the transition preserves the total while changing the meaning carried by individual records. I want arithmetic reconciliation to remain rigorous and to sit beside an equally explicit account of customer instructions. The strongest release recommendation explains both forms of evidence and the exceptions still held. That gives management a basis for judgement more substantial than a matching balance and a confident launch announcement.