Choose the record set and the reason
Start with a specific operational problem: duplicate customer entries, missing supplier references or inconsistent product attributes. Define the system, fields and date range involved. A bounded cleanup produces a reviewable result; an instruction to fix all company data does not.
Preserve the source before proposing changes
Keep an approved snapshot or export and record when it was taken. Work in an authorized staging copy where possible. Establish which system governs each field and who can approve a change. Never assume that the newest-looking value is correct.
Classify the issues
Use a small issue register that distinguishes confirmed formatting corrections from uncertain record relationships.
| Issue | Preparation work | Hold for review |
|---|---|---|
| Inconsistent format | Apply an agreed field rule in a draft | A conversion changes meaning |
| Missing value | Locate the approved source | No supporting source exists |
| Possible duplicate | Compare identifiers and linked records | A merge could lose history |
| Conflicting value | Preserve both values and source dates | The governing source is unclear |
Example: two customer names look alike
Two rows may share a similar business name but belong to different branches. Compare stable identifiers, source records and linked activity before proposing a merge. In this illustrative case, the assistant records a possible match and the evidence; the owner decides whether the records describe the same entity. Similar spelling alone is insufficient.
Review a sample before applying a batch
Ask the owner to review routine corrections and ambiguous cases. Record the original value, proposed value, rule and reviewer response. Resolve unclear rules before expanding the batch. Use the system’s approved recovery method rather than assuming every change can be undone.
Reconcile the result
Compare the approved scope with the changes actually made. Check record counts and the fields affected, retain rejected items and report unresolved cases separately. A shorter spreadsheet is not evidence that duplicate relationships were resolved correctly.
Keep the cleanup from recurring
Feed common errors back into field definitions, intake instructions and review checks. Decide who maintains the rules and when a new exception should be escalated. Compare data entry support with document management, or discuss a bounded cleanup queue.