What decreases efficiency and impairs the learning of my AI?
The #1 reason that the AI learns slowly - or even incorrectly(!) - is that predictions are changed by arbitrary edits.
Arbitrary edits are changing values to those not found on an invoice. By making these changes, you are confusing the AI as you're telling it the predictions are wrong, negatively affecting your automation efforts.
Common arbitrary edits include:
Adding text to predictions: Commonly happens to invoice numbers, by appending the invoice date or vendor acronym to the invoice number.
E.g. "123456" to "Inv. 123456"
Changing the invoice date/ due date to a different value not found on the invoice. This may happen when clients wish invoices to be paid on a cash accrual basis.
Removing predicted fields: Commonly happens to Payment Terms, removing a predicted field tells the AI it was wrong.
Deleting predicted line items: If the amounts are correct, and creating new ones, instead of updating GL account/dimension coding to correct coding.
What do I do about Utility/Recurring Bills without an invoice number?
For invoices that do not have an invoice number, Vic.ai will instead predict the invoice number as a combination of fields, namely the predicted Account Number and Invoice Date.
The format will be specifically: AccountNumber-mmddyyyy
For instance, an account number of 12345678 and Invoice Date of July 4, 2024 would be 12345678-07042024
The AI will usually substitute a Customer Number or similar unique identifier if Account Number is not found.