Preserve what each file actually contains
Two imports can look alike without representing the same set of events. Keep the original files untouched and record their account identifiers, stated date ranges and download dates. A later download may contain corrections or additional posted transactions. Its newer timestamp does not establish that every older row should be discarded.
If two files are byte-for-byte identical, that establishes that the files are copies, but it does not by itself explain how many times their rows entered the working record. Trace the import batches as well as the files. An audit needs to distinguish duplicate source material from repeated processing of a source that was downloaded only once.
Compare identifiers before descriptions
Start with the strongest event identifiers available in the documents. Account identity and debit or credit direction matter alongside transaction references. Merchant text is useful supporting evidence, but a processor description can recur across several genuine purchases. A matching amount and familiar name are a reason to investigate, not a complete duplicate test.
Keep the source values visible when a working comparison normalises spacing or punctuation. Removing formatting from a reference can help comparison, but removing meaningful leading characters might merge distinct identifiers. Document any transformation used. A reviewer should be able to recover the original value rather than being asked to trust an unexplained cleaned field.
Treat dates as named facts
A purchase date, posting date and export timestamp answer different questions. If one file provides the purchase date and another provides the posting date, compare the event sequence rather than forcing the dates to equality. Retain both labels where available and look for supporting receipt or reference evidence.
Pay particular attention to date formats when combining exports. A value such as 04/05 is ambiguous without the source's stated convention or wider context. Do not resolve it by choosing whichever interpretation produces the most matches. Preserve the ambiguous raw value and establish the relevant format before using the date in a duplicate decision.
Examine an ambiguous pair
Suppose, illustratively, file A shows a 480-unit debit on 6 November with reference K17. File B shows 480 on 7 November, also with K17, for the same account. If the documents identify the dates as purchase and posting dates for that event, one retained payment can carry both source references. The one-day difference need not create a second expense.
Now suppose file B instead shows reference K18 and a separate receipt supports another 480-unit purchase. The repeated amount is genuine spending, and both events belong in the total of 960. The correct decision depends on evidence that distinguishes the two cases, not on whether a lower report total seems more plausible.
Explain missing and conflicting evidence
Where references are absent, compare the full available context and record the remaining limitation. A nearby account balance or a receipt can help, but do not claim that a probable match is certain. Maintain an unresolved group if the available documents do not establish whether one or two payments occurred.
If sources conflict, retain both interpretations until the conflict has a factual resolution. For example, a receipt may establish a purchase while an export omits its final debit. That omission is not proof that the purchase never occurred. Equally, a receipt alone should not establish that a particular account was charged when it identifies another payment method.
Record a reviewable match decision
For each excluded duplicate, preserve the retained event, the excluded source row and a short explanation of the evidence supporting the match. Include the account and relevant reference rather than relying on row position alone. Row numbers can change when a spreadsheet is sorted or a file is exported again.
Finish by reviewing a sample of retained repeated amounts as well as removed pairs. This checks whether the method protects genuine repeated purchases instead of merely finding obvious overlap. A useful evidence review produces a transparent set of decisions, an explicit unresolved list and enough source detail for another person to reproduce the conclusions without guessing how the files were interpreted.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


