Kurzer Check
Guess first, no penalty. You ask an AI to "remove duplicates" from a customer list, and 340 of 5,000 rows disappear. What should have come first?
Cleaning means fixing messy data: extra spaces, mixed date formats, blanks and repeats. AI is quick at it. But every cleaning step is a decision about what to keep, and those decisions change your totals.
So write the rules before the AI touches anything:
- Duplicates. Which columns must match for two rows to be the same?
- Blanks. Leave them empty, fill them, or drop the row? In SQL, AVG and COUNT(column) skip blanks (NULLs) without telling you, while COUNT(*) counts every row.
- Formats. Which date format is the source in? 03/04 is 3 April in one country and 4 March in another.
Then keep the raw file untouched, and count rows before and after each step.