How to Clean Data Without a Data Engineer

Cleaning is not a phase, it is a set of promises about your data that you can check. Written as checks, cleaning stops being an endless chore and becomes a short, repeatable script.

Begin with structural promises: one row per what, which fields must never be null, which must be unique, and what ranges are plausible. Run those checks first and count the violations rather than silently dropping them.

Then handle the four recurring offenders: duplicate records from repeated intake, inconsistent categories from free-text entry, dates stored in mixed formats, and identifiers that change between systems. Each deserves an explicit rule recorded in a decision log, not an ad hoc fix in a cell.

Never edit the raw file. Keep the original untouched, write transformations in a script, and output a dated clean file. When someone asks in six months why the numbers moved, you will have an answer instead of a guess.

Finally, publish a short data quality note alongside every analysis: rows in, rows excluded, and why. It builds more trust with boards and funders than a polished chart ever does.