Clean the fields needed to identify records and run the first workflow. Do not spend months perfecting historical data that the initial release does not need.
Prepared with AI assistance. These are practical scoping recommendations; examples are illustrative, not client results.
Define the target use
List the fields that support matching, assignment, customer contact, and the first reports. Inspect representative files to see how those fields are actually used. A column that looks optional may carry essential information in informal notes, so ask the people who maintain the spreadsheet before removing it.
Classify problems
Separate duplicates, missing values, inconsistent formats, and conflicting definitions. Give each class an owner and a correction rule. Do not let a developer guess the correct customer or status from context when the business needs to decide. Keep uncertain records in a review list rather than manufacturing plausible values.
Preserve the original
Retain a controlled source copy and record how cleaned data was produced. This allows the team to investigate a surprising imported value. Avoid repeated manual edits across several copies with no clear final version. A repeatable cleaning process is more useful than a one-time file nobody can reconstruct.
Check against the workflow
Load a trial sample and ask users to complete real actions with it. Correct formatting is not enough if the records still lack the information needed for assignment or follow-up. Decide which remaining issues must block launch and which can be handled through a visible, owned cleanup process afterward.