Consistent categories are agreed labels for grouping records. They make comparisons possible without requiring someone to reinterpret every entry. Consistency involves both spelling and meaning: two identical labels can still describe different things.
Keep the list small enough to use, explain the boundaries between categories and include a way to mark uncertainty. Preserve original values when standardising existing records. A mapping from original to agreed labels gives you a way to inspect and reverse questionable decisions.
A small-business example
A repair workshop records job states as “Done”, “Complete”, “Finished” and “Collected”. The first three may mean the work is finished, but “Collected” records a later event: the customer has received the item.
Replacing all four with “Complete” would make the column look cleaner while erasing a useful distinction. The workshop instead defines repair status and collection status separately, then reviews older entries where the wording does not reveal which event occurred.
Try this
List the distinct values in one frequently used category column. Write a definition for each proposed standard label. Map obvious spelling variants, but leave ambiguous values in a review list. Ask the people who enter the records whether the proposed categories match the work they actually do.
Record the choices in a Data dictionary, use Human review for uncertain mappings and keep the transformation trail through Provenance.