LEARN. PRACTICE. GROW.1. Separate the future state from the transition
The future solution should support querying reconciled positions. Reaching that state may require historical conversion, training, temporary coexistence and a return procedure. These transition requirements have their own lifecycle: they do not disappear because the new feature passes tests. Identify when each temporary capability is needed and its retirement criterion. Data comparison should begin with a contract covering included population, business date, transaction state, keys, currency, units and authorized transformations. A global count without these boundaries may compare different populations. Use synthetic data to develop the method, then confirm real-data quality and access in an authorized environment before relying on the method for acceptance.
2. Recognize false success from totals
In the fictional worksheet, the source contains two EUR positions of 100 and 250 and one USD position of 150. The target retains three rows and the same currency totals but replaces the EUR 100 position with another reference. Count and sum checks pass although one position is missing and another is unexpected. Compare identity under the contract first, then relevant attributes, values and states. Do not add EUR and USD into one total without an explicit conversion rule and valuation time. Nor should every reference be uppercased merely to increase matches when the contract distinguishes case. Comparison rules need agreement from information consumers and checks using normal and exceptional examples.
3. Address duplicates and differences before aggregation
A repeated key should not silently be replaced by the last row read. It may indicate improper duplication, incorrectly defined granularity or a legitimate multiple-movement rule. Record the conflict and confirm the contract before choosing deduplication or aggregation. If amounts use integer minor units, make the unit explicit; comparing 100 with 100.00 without knowing the unit can hide a scale error. A zero net difference can also hide opposing differences across accounts. The local exercise identifies missing, unexpected and duplicate keys and per-record amount differences, retaining currency totals as a complementary control. Its output identifies investigation candidates rather than automatically approving a migration or deciding which system is correct.
4. Connect criteria, exceptions and authority
Define criteria before cutover and link each to the requirement, dataset, test version and acceptance owner. An unexecuted test is unknown; it is neither implicit approval nor proven functional failure. An accepted exception should identify who can accept the risk, scope, expiry, compensating measure and closure condition. Closing a ticket does not automatically make its operator the change approver. If a fix changes a data transformation, assess dependent tests and consumers rather than repeating only the case that exposed the defect. At the committee, show each criterion’s state and the gaps conditioning cutover, extended coexistence or return. Keep the recommendation separate from the authority required to approve the chosen option.
5. Check operation, adoption and legacy retirement
A migration can reconcile correctly and still fail to deliver expected value if the night shift lacks access, training or support procedure. Distinguish solution defects from organizational or process limits restricting use. Before retiring legacy systems, confirm consumers, required history, retention, support and the agreed query or recovery route. An old backup alone is not an accepted historical-query service. Measure adoption and outcomes using stable population, period and denominator while retaining differences between teams. Do not attribute causality to the change merely because an aggregate indicator improved. Handover should make criteria operational: who observes, who decides, when they intervene and what evidence they retain during stabilization.
Three rows and equal totals by currency can hide a missing reference and an improper extra one. Reconciliation should compare identity and content, not only aggregates.
Common pitfalls
Adding currencies without a contract, deduplicating without a rule, accepting unknown tests, confusing a closed ticket with approval and retiring legacy before confirming consumers.
Related topics: Decisions, versions and dependencies · Model, verify and validate requirements · Measure outcomes and interpret evidence
Accept transition through identity, explained differences, verifiable criteria and prepared operations.
Reference: The Business Analysis Standard · CBAP six-knowledge-area blueprint, May 2026 handbook