AI and Data Normalization: Can ChatGPT Help Us Clean Up Data?

One of the biggest recurring headaches for finance teams is getting raw data—especially from banks—into a clean, standardized format that ERPs like NetSuite will accept. This isn’t glamorous work, but it’s foundational: if the data doesn’t import cleanly, reconciliations, reporting, and automation all grind to a halt.

Today I tested whether AI could help ease that burden by trying to convert raw bank statements into a NetSuite-compatible CSV format.

The Test: From Bank Statement to ERP-Ready Format

We started simple: a U.S. bank statement downloaded in CSV format. These files already resemble NetSuite templates fairly closely, so this should’ve been a layup.

Gemini's Attempt:I tried Google’s Gemini model first. Instead of parsing the file, it gave me a 1,100-word lecture on the risks of handling financial data. It refused to proceed due to “security and privacy risks.” So—dead end.

ChatGPT’s Turn:ChatGPT, on the other hand, took a far more helpful approach. I uploaded the CSV and it quickly analyzed the structure, understood the column meanings, and generated a cleaned-up version.

There were a couple of glitches:

• It didn’t catch that “Check” wasn’t a valid transaction type for NetSuite.

• In some cases, it appended random zeros to the memo field.

Still, overall it did what I asked—rename headers and prep the file for import.

Next Challenge: Israeli Bank Statements (in English and Hebrew)

Then came the real test: non-U.S. statements.

Israeli Bank Statement (English):ChatGPT handled this surprisingly well. It correctly identified where the transactions began, and interpreted positive/negative values as credits and debits.

Hebrew Bank Statement:Here I was impressed. ChatGPT explained its translation logic, handled local date formats (DD/MM/YY vs. MM/DD/YY), and produced a file close to usable.The only issue? It scrambled the Hebrew memo fields into gibberish.

When I pointed this out, it fixed them—but declined to translate them to English, citing the need for integration with services like Google Translate or DeepL.

Final Thoughts

Gemini simply refused to help. ChatGPT passed the basic test and shows real promise. With a few tweaks—and perhaps a layered approach using translation tools—ChatGPT could genuinely streamline a part of finance work that’s usually time-consuming and manual.

I didn’t go as far as trying to match payees with customer/vendor names in NetSuite yet. But that’s the next frontier—and if AI can solve that one, we’re talking serious time savings.