We have already described automated invoice processing based on workflows here and here, and have also set it up for numerous customers. However, in this scenario, the invoice data such as customer number, invoice amount, etc. must be entered manually into corresponding forms in the system. An AI such as Microsoft Syntex, on the other hand, is able to process corresponding documents independently and recognize the metadata.
For this task, the AI must be “trained”, using sample documents. At least 5 files of the desired document type must be uploaded as positive examples, and one file of a different type as a negative example or counter sample.
In the next step, the uploaded positive examples are given a field label, and the recognized values in the document are confirmed or rejected so that they can be identified with certainty.
The values extracted in the learning process result in a new content type that can be applied to a document library to identify the files uploaded in it as invoices, extract the relevant information (customer number, invoice number, invoice total, etc.) and save it to the document along with the content type. This library can also be created “empty” in advance, because as soon as the model is assigned to this library, it expands it with the corresponding columns of the content type.
A separate mailbox can be set up for invoices received by mail. Everything that ends up there as an attachment to a mail is stored in the document library for invoices and analyzed and processed accordingly. Documents whose properties differ from the specified ones are assigned the library’s default content type after analysis, so that they can be filtered and checked.
The Forms Designer supports the setup of different content types, so the appropriately stored forms will be loaded only for the correct content type.
The business logic from the invoice processing mentioned above still applies, e.g. that an approval must be obtained above a certain invoice amount.
Microsoft Syntex can be a useful addition and take over work steps that were previously performed manually. We would be happy to show you a small demo and accompany you on your way to AI-supported invoice processing!