Linguistics

It's usually clear from context, and also allows for some hilarious number puns.

I believe 499 would have been written CDXCIX. 994 should be written CMXCIV. You only use a single letter prefix to subtract (hence VIii not IIX) and you only use the two symbols prior to the one you're subtracting from, so you can subtract C or D from M but not X or VI.

This kept complexity of parsing down but also produced fewer forms that could be confused with real words!

Besides context, like Nemo said, sequences of letters that are legit for numbers are often invalid for common words. For example, in Latin you'll never see words with "MC" (because */mk/ is phonotactically forbidden), and yet you'll see it for plenty numbers. Or "III", it pops up all the time for numbers but almost never for actual words.

On the conlanging part: if you're using an alphabet, and your phonotactics prevent consonant-only words (pretty common restriction), you can ensure the numbers are obvious as numbers by using only consonants. Another alternative would be to create one grapheme to prefix numbers with; like, instead of writing "ID" you'd write "#ID".

There isn't really a scientific measure of how confusing something is linguistically that I am aware of. As you and other people have pointed out, there are plenty of languages that don't differentiate orthographically between numbers and letters. It's like asking whether people will confuse read and read. Yes, they probably will in some cases, but on some level speakers understand that these words are written identically and that you require context to figure it out. So they would know the difference much more often than not

You might follow the pattern of braille and have a prefix symbol that indicates the following letters are numbers. ⠼ functions like a number or hashtag symbol in front of letters A - J to indicate they're to be read as numbers, 1 - 9 & 0, not letters.