Automation of Vendor Invoice processes in LINK

Summary: AI being capable to recognize the need to create a new partner and distribution based on a incoming invoice document and process it without human intervention.

Currently, we have different processes for integrating electronic invoices from vendors (PurchaseInvoice) in several countries as part of our support in LINK Portal. Examples include FINVOICE, IDOC PEPPOL, Italy Sales Invoices, or UBLXML Invoices. There are internal guidelines on our company on how to set up these processes.

This means that every time a new vendor provides an invoice via LINK and the partner does not already exist, the document fails in the LINK portal and gets assigned to our first-level support queue.

Information that is needed in general for LINK to process the document includes:

• Distinguishing between a Sales Invoice or a Purchase Invoice.

• Recognizing the input syntax and format of the document.

• Assigning a Sender Party to a sender partner in LINK based on a specific identifier (CVR, GLN, VAT, etc.).

• Assigning a Delivery or Receiver Party to a receiver partner in LINK as a receiver of the invoice, also based on a specific identifier (mostly receiver GLN).

• Setting a delivery endpoint according to the final destination of the document and its purpose.

• Assigning an output format in which to convert the document in case it is needed based on the final endpoint.

Most failures at the first-level support are caused by the receiver partner not being found or due to not existing or not being recognized with the given identifier on the invoice (CVR/GLN). Logically, if the partner does not exist, there would be no distributions (receiver; formats in/out; endpoint) assigned.

This task is currently done by human examination of the incoming document, which visually checks and collects all information needed as explained in the steps above and proceeds to create a new partner and distributions according to the given details and our guidelines. It is remarkable that no validation of the content of the invoices is done on this manual process. As far as the invoice can be processed on the created distribution with the given mapping, the document gets completed and delivered to the final endpoint for a final validation as part of a finance process.

Therefore, human analysis of the invoices when failed does not provide much value to the overall process of receiving electronic invoices from vendors beyond assigning the right setup for them to be delivered to the right endpoint by creating the needed record for a partner and distribution on LINK.

As this is determined by our internal guidelines, it should be a straightforward decision in most cases to decide which setup is needed. In case the document fails because the partner does not exist, we should decide whether to create a partner and its expected distribution just based on the format and sender/receiver information contained on a given document.

If an AI or automation were capable of recognizing the given input parameters to create a new partner and distribution without human intervention and to process the document in LINK according to the expected mapping to the final endpoint, this manual task of processing failed vendor invoices due to not having received any previous invoice from an unknown vendor would be skipped and processed by the system. The final decision of validating if the content of the invoice has been processed correctly would be left to a further step as part of a finance process and not a technical one. Of course, in case the automation could not simply process the invoice by deciding which is the needed setup or for whatever other reason, human intervention would be needed by providing value in examining the reason why it can not be processed.

One of the keys would be the capability of the automation to identify the sender and receiver of the invoice based on a specific identifier parameter (GLN/CVR/VAT/FI:OVT…) and to decide when it is needed to create a new partner without duplication of partners. We have already tested if AI like ChatGPT is capable of recognizing for example an XML UBL purchase invoice, finding sender, receiver and identifiers succesfully.

Currently, in our particular circumstances only a few vendor invoices per day are failing in the portal due to the need for this process of manual onboarding of new vendors into the process. However, in view of the future needs for the implementation of electronic invoices in several countries, there is an expectation to face an increase in the need for these integrations.