How do track errors with the AI? Is it up to me as a user to flag them or do you have a review process or have different models confirm the same outcome?
Right now if you suspect an error in AI response you can flag it/submit as feedback. For anything that AI does without your input (bookkeeping) we associate a confidence with it and when below that threshold we get a human in the loop.
We know that trusting AI with financial info is hard which is partially why we have the sheet (so that the AI can backup it's claims).