Inspyrus Enhanced Invoice Extraction
By: Karla Broadrick | Technical Architect, WebCenter Imaging Team
Oracle WebCenter Forms Recognition is not a tool that is unique to the Inspyrus Solution. WebCenter Forms Recognition (WFR) is a tool that has been around for many years. WFR is a key component of how Oracle enables accounts payable automation because it provides intelligent classification and extraction of invoice metadata. Key fields such as Invoice Date, Invoice Number, Supplier, PO Number as well as line items are intelligently recognized for a large number of suppliers based on a pre-trained set of invoices.
The Inspyrus solution utilizes WFR in a “black box mode.” This means that invoices pass through WFR where the invoice metadata is extracted but instead of being routed to the Verifier application for correction, invoices do not stop in WFR but instead go directly to the Inspyrus UI. It is there that invoices are able to be corrected, if needed, before being routed to the applicable ERP system.
What makes the Inspyrus solution’s use of WFR different than traditional solutions is that the out of the box WFR product has been enhanced to alleviate common problems and shortcomings experienced by a large number of WFR customers. With these enhanced invoice extraction capabilities, the Inspyrus solution brings invoice automation to the next level, one of the many reasons the solution is considered the next generation in accounts payable automation.
One of the most basic examples that long time users of the WFR product can relate to is the extraction of the PO Number field. The way that PO Number extraction typically works is that any number of PO masks can be configured in the INI file. The masks can be as specific (TEK#####) or generic (6 digit number) as needed. As a best practice, I advise customers to use PO masks that are as specific as possible.
If masks can be kept very specific and the number of them kept low, WFR typically does a pretty good job of finding the PO Number on the invoice, regardless of its location. After the PO Number is extracted, it is then validated against the ERP database to ensure that it is a valid PO. If it is not found in the database it is flagged as invalid and will require user correction.
Where problems arise is when business needs dictate that PO masks must be generic (######) or a large number of masks are required (as often happens in larger companies with multiple systems generating PO’s). In these cases, WFR often does not do as good a job of correctly choosing the actual PO number. Despite clear keywords such as “Purchase Order Number” that could contextually help locate the actual PO number, WFR will often incorrectly choose another value on the invoice that matches a mask but is nowhere near any labels that would indicate a PO Number.
The Inspyrus WFR solution addresses this issue by offering a routine that enhances the out of the box invoice automation by using contextual evidence to help identify the correct PO number from the possible candidates.
TekStream recently worked with a customer that faced both of these challenges. The screen shots below illustrate the difference between the out of the box WFR extraction and the extraction results with the Inspyrus solution. This is just one of the many examples of ways the Inspyrus solution takes accounts payables invoice automation to the next level.
Out of the box extraction: PO Number is chosen from the text under a barcode on the page that matches one of the many masks. The number chosen has no context to suggest it is a PO number.
Inspyrus extraction: PO Number is correctly chosen as the candidate following the label “P.O. #”