Cleaning is where you fix what can be fixed automatically and escalate everything else to the client. The output is a source that the target system will accept, and an exception list where every row has a decision. This article covers the split between the two, and how to de-duplicate without destroying real records.
The short version. If a rule can fix it safely, fix it. If a human has to decide, it goes on the exception list, and the list is a gate that closes only when every row has an answer.
What a rule can fix
Which of these do you think is riskier to automate: trimming spaces, or standardizing a date? Have an answer before the list.
Safe to fix by rule, because the intent is not in doubt: leading and trailing spaces, inconsistent capitalization in codes, a date format that is unambiguous, a phone number written with different punctuation, a country name that appears in two spellings.
