I have philosophical objections to this kind of thing.
1. SSN's should be redacted from transcripts anyway, they should *not* appear as plain text in the body of the file.
2. SSN's (and strings of numbers of numbers) are hard to validate -- in part because they were invented before checksums became as commonplace as they are today -- unlike credit card numbers.
3. In the absence of checksums in social security numbers, birthdays, etc., the next best thing is to transcribe exactly as dictated. If someone is failing to make it clear that they are giving you a price, or social security number, or birthday, or whatever, what else might they be screwing up?? It's not really our place, as transcribers/reporters to be making those kinds of assumptions. It's a slippery slope from this to correcting perceived "mistakes." For instance, when I transcribed insurance claims memos, I was ruthless around the beginning of the year when people got the year wrong. If it was January 3, 2013 and someone mistakenly dictated 2012 -- tough! I'd rather force the letter to be redone by the person dictating than to have it go out to a policyholder with, say, a claim amount off by an order of magnitude.
4. It's duplicative. There are lots of applications -- database software, spreadsheets, word processors, etc, that are quite capable of formatting strings and sequences of numbers. No need for this at the raw transcript level.