Fun fact: I did. And not just for accounting systems, but all sorts of data ingestion pipelines. Did it work every time? No. Did it work in many cases? Yes. Is that better? Absolutely.
Here is the thing: If I accept broken CSV, where do I stop? What's next? Next thing my webservice backends have to accept broken HTTP? My JSON-RPC backends have to accept JSON with /*/ style block comments? My ODBC load-balancer has to accept natural language instead of SQL statements (I mean, its the age of the LLM, I could make that possible).