I used to open a messy dataset and immediately start fixing things.
Remove duplicates. Correct spellings. Fill blanks. Standardise formats.
It felt productive.
Then I realised I was sometimes fixing problems I hadn’t even properly understood.
Now, before touching a dataset, I do a quick audit.
How many duplicates are there? How many missing values? Which formats are inconsistent? Are different categories actually describing the same thing?
That small step changes everything.
A few inconsistent entries might be a five-minute cleanup. But if hundreds of records are inconsistent because different people have been collecting information differently, the problem isn’t really the spreadsheet.
It’s the process behind it.
I also stopped assuming that manual checking is automatically more accurate. Going through hundreds of rows by hand at 2am doesn’t make me careful. It makes me tired.
If a formula, validation rule, or script can reliably identify the problem, I’d rather automate that part and use my attention where it’s actually needed.
The goal of data cleaning isn’t to make a spreadsheet look perfect.
It’s to make the information trustworthy enough to use.