For about the last week and a half, one of our users will notify me that when they go in to process invoices through IDC, lots of fields will be blank. I’ll go in to check and sure enough, invoices for vendors we’ve been dealing with for months are coming in and it’s not grabbing the OCR or many of the other fields. I went into our server and restarted some of the services associated with Epicor IDC, then restarted those batches from within IDC and it appeared to fix itself.
Then the next day, it would happen again. And again the next day. For some invoices, the restarting doesn’t even help and we will just end up clicking on those fields manually. Has anyone else run into an issue similar to this?
Issues with IDC usually are training data related or DFD handling. I would encourage people to upgrade to latest version 9.34, enable azure ocr and this may help as well.
Possibly your services need to be restarted daily with a scheduled task. Cloud IDC is really good with azure enabled by default.
I’d love if you find a solution for this - our customers are having the same thing.
Support only says it’s to do with “inconsistencies” with training, but our customer is just putting in one document type (invoices). There’s no real ambiguity with the fields.
I find that sometimes a user can train a document multiple times and have the same issues with training. Sometimes the solution is to delete the training data for the supplier and retrain the document.
When IDC works well it is usually Azure OCR engine, the dfd has the keywords/exclude values for the supplier, and the training was done in a manner that IDC can repeat (no typing in data capture field). Along with all of these steps, be prepared to repeat training the same document multiple times to validate the training.
If a supplier document is poor and the fields overlap, then you could create a localization on the DFD to set certain formulas to occur if the VendorName = Example Vendor. If the field data that you capture is in incorrect formats, you may need a different document type with language locale. I would advise to try one document and see if any fields on the document are captured, add a keyword and repeat training to see if the data capture improves.