We use the Document Understanding Invoice model along with AI fabric for processing invoices automatically, which works great. 99% of the invoices that we automatically process based on high confidence in all fields seem to be accurate and well. Recently, accounts payable shared an example invoice with an issue related to grabbing negative values.
We have a vendor that sends negative values in credit memos that have parentheses “(100.00)” around the number that is not being recognized as negative and I am unsure of how to deal with this.
Our invoices are being parsed with OCR, then put through the Machine Learning Extractor using an ML Skill that we have retrained with the Invoices Document Understanding model. This model works very well for getting accurate values on most of our vendors and invoices. We have a job in one of our Workflows that looks at the values returned from the ExtractionResults from the Machine Learning Extractor and automatically processes the invoice into our accounting system if all of the values have high confidence on all required fields. This has been working great to save everyone time in accounting. I am a bit lost on how we can change how the Invoice model interprets negative numbers though. It does recognize negative values with a minus (-), but does not with parathesis (()). This creates an issue with how we automatically process invoices extracted with high confidence because they get entered as Invoices rather than Credit Memos. I have not been able to find anything in the documentation to help.
Is there some kind of setting or easy way that I can set up the Invoice model to recognize parenthesis surrounded numbers as negative?
Here is an example (This is in validation station just to show the base results, but this invoice was automatically processed because each field has high confidence and values found, even though the negative value was recognized as positive):