I've been using hledger for two years now, and I don't use any exported data from my banks. Here's my workflow:

Everytime I make a transaction, I put a note on my phone. The note is a simplified version of hledger file format. Example:

24feb 12.5 cash food eating out at xx 6.5 bank1 phonebill 25feb ...

and so on. As you can see, each block is started with the date, and each line corresponds to a journal entry. In each entry, the first item is the dollar amount, the 2nd item is the credited account, the 3rd item is the debited account, and the rest will be taken as the entry description.

Every week end, I will parse this note with a Python script that I wrote, and put the output in the actual journal file. The shorthand account names in the note will be converted into actual account name (e.g. 'bank1' to 'assets:bank:bank1'), based on a dictionary file.

I'd then proceed to manually check that all ending balances in hledger match with the actual amount in the real world. (It's pretty satisfying to see the numbers match.)

After I've finished processing the note from my phone, that note will be archvied and I will start with a new note.