The string “1,250” can mean one thousand two hundred and fifty, or one and a quarter. If an AI combines figures from several documents, a tidy output column can conceal that ambiguity. Decide what the source meant before calculating anything.
This is a question about interpreting written numbers, not converting their units. Changing kilograms to grams will not repair a value that was already read a thousand times too large.
Keep the original beside the converted value
Work on a copy. Ask for separate columns containing the original string, source, stated number convention and proposed numeric value. If the convention is unknown, the proposed value should remain unresolved rather than silently becoming zero.
Use a small example with explicit rules: source A defines a decimal point and gives “12.50”; source B defines a decimal comma and gives “12,50”. Both mean twelve and a half, so their sum is 25. An additional “1,250” with no source convention stays outside the total until clarified.
Do not infer every row’s format from the document’s language when it combines exports from several systems. Ask the person who supplied an unclear value, or check the original export specification.
Separate interpretation from display
Excel can use regional decimal and thousands separators. Power Query also uses locale when converting text into data types. Those are settings to understand, not reasons to replace every comma in a mixed file.
Group rows by a confirmed source convention before conversion, then check a few values and subtotals. Keep the original strings intact. Finally, verify that changing the display format has not merely made an incorrect number look more familiar.