I believe that's where "natively" comes in. If you store it in binary, you can operate on the values as numbers. If you do some decimal conversion, it's a blob of data. (Yes, there are BCD instructions but they're massively limited)

Decimal numbers are still "stored as binary" at the silicon level.

I meant as opposed to floats, precision loss, etc.