Review fuzzy supplier matches before changing finance records

A supplier list imported from another system can contain abbreviations, spelling differences and trading names that do not exactly match the approved vendor master. Fuzzy matching can help an analyst find possible relationships. It should not establish the payee, merge supplier balances or authorize a change to payment details merely because two names resemble one another.

Microsoft describes Power Query fuzzy merge as an operation on text columns, with options for similarity thresholds, displayed scores and the number of returned matches. Those settings concern text comparison. They do not verify that two records belong to the same legal business or that an invoice should be paid to the account on a suggested record.

Define the intended output before running the merge. A review queue containing candidates and evidence is different from an approved crosswalk used to post transactions. Keep that distinction visible in file names, column labels and permissions so an exploratory result cannot be mistaken for the final mapping.

Inspect the source for a stable supplier reference already shared with the destination. A maintained account code or approved migration crosswalk may resolve a relationship without approximate name comparison. Preserve the source system and company context with the identifier, because a code that is unique inside one company may occur elsewhere with a different meaning.

For records that still need review, retain the original name alongside any normalized comparison text. Removing punctuation, shortening a name or standardizing a suffix can make candidates easier to find, but those operations can also remove a distinction between businesses. The reviewer needs to see what the source actually said, not only the cleaned label that produced a strong score.

Gather appropriate supporting fields without collecting unnecessary sensitive data. An existing customer or supplier number, documented trading-name relationship, address and relevant contract reference can help investigate identity. A matching postal address alone may be shared by several businesses. Treat corroboration as evidence to assess rather than a rule that automatically overrides an inconsistent identifier.

Review the match settings and preserve them with the run. Microsoft allows a maximum number of candidate rows for each input and provides an option to display similarity scores. Limiting the output to one candidate does not prove that no other plausible supplier exists. The limit can remove the very ambiguity that a reviewer needs to see.

Use an invented example: a source says Riverton Services, while the master contains Riverton Services North and Riverton Services South. If both appear plausible, a highest-ranked result is still only a candidate. Ask which agreement or existing account relationship identifies the intended business. Leave the record unresolved when that evidence is unavailable rather than selecting a destination to make the unmatched count fall.

Keep the candidate score next to the comparison method and the source values. Do not present a similarity score as a measured probability that the supplier identity is correct. Changing case handling or a transformation table can change the matching behavior. The operating team should know which settings produced the candidates it is being asked to review.

Give the reviewer clear outcomes: approved relationship, rejected candidate, several candidates requiring evidence, or no suitable candidate. Record the chosen destination identifier and the evidence supporting an approval. A comment saying looks right is difficult for another person to verify and does not explain a disagreement between the source and master data.

In a hypothetical queue of 60 unmatched source suppliers, review establishes 38 relationships, identifies 12 ambiguous cases, finds no suitable candidate for five and leaves five waiting for documents. Only 38 links are ready for the approved mapping. The other 22 remain visible with their next action. Reporting all 60 as matched because the software returned something would overstate completion.

An approved identity link does not authorize every change that could follow from it. Combining duplicate master records, reallocating invoices and updating payment instructions can require separate business review. State which operations the mapping may support. Retain existing payment and accounting controls instead of letting a data-cleaning exercise quietly acquire authority over transactions.

Before using approved links in a report or migration, check whether each source key resolves to the intended number of destination records. Expanding several candidates into transaction rows can multiply an amount. Compare the transaction population and totals before and after applying the accepted crosswalk, with ambiguous cases excluded from automatic application and separately reported.

Test a source supplier that has changed its trading name, two unrelated suppliers with similar names, an inactive master record and a record with no evidence of identity. Include a correction to a previously approved link. The expected behavior should describe both the relationship and the treatment of transactions already processed under the earlier mapping.

If a mapping was wrong, identify the outputs that used it before changing history. A supplier total in a management report, a draft import and a posted transaction need different correction routes. Preserve the old mapping version and the approved replacement so the accounting owner can trace which records require review.

A transformation table can help with known name variations, but it should represent reviewed relationships. Microsoft permits custom mappings as part of fuzzy matching. Keep those mappings scoped to the relevant source and purpose, with an owner and supporting evidence. An abbreviation accepted for one supplier population may be ambiguous in another company or country.

Review new unmatched names and material source changes without reopening every established relationship. Retain unresolved cases instead of repeatedly lowering the similarity threshold until they disappear. The useful result is a maintainable crosswalk with known limits: finance can explain which links were approved, which still need evidence and why the next import can rely on the accepted subset.