Data cleaning is the work of finding and fixing problems in a dataset so it can actually be trusted. That means correcting typos, filling or flagging blanks, standardising formats and removing records that are wrong or duplicated. The goal is data that reflects reality rather than the mess of how it was collected.
Imagine a contact list where one person is ‘NL’, another is ‘Netherlands’ and a third is ‘the netherlands’. To a computer those are three different countries. Cleaning rewrites them to one agreed value, the same way you would tidy a shared spreadsheet before sending it to a client. It pairs closely with deduplication to merge repeated records and with data validation to stop bad values getting in next time.
Format mismatches are just as common as typos. Dates that arrive as 03/04 are ambiguous until you know whether the source uses day-month or month-day, and a phone number stored with spaces, dashes and a country code in random combinations will never match cleanly. Part of cleaning is picking one canonical shape and rewriting everything to it, so later steps can compare values without guessing.
In practice cleaning is rarely a one-off. New data keeps flowing in, so the rules usually live inside an ETL process and run every time fresh records arrive. Done once and forgotten, a dataset drifts back into chaos within weeks. The aim is not perfection but a known, repeatable standard everyone downstream can rely on. A good rule is to fix problems at the source where you can, so the same typo does not need correcting on every single run.
At TopDevs we treat data cleaning as a continuous step in the pipeline, not a manual rescue mission a client has to repeat by hand.
