Document classification

Document classification is the automatic assignment of an incoming document to a document type, for example an invoice, credit note or reminder. Based on this assignment, the software determines which document is created and which workflow is started.

Functionality

AI-based classification is based on machine learning. Special algorithms learn a model from example data. They then apply this model to new, previously unseen data.

One way to achieve this is supervised learning: The learning algorithm receives not only the raw data but also the expected result. If the output of the model deviates from this, the algorithm adjusts the model accordingly. The basis for this is labeling. Labeling refers to the labeling of data with additional information, such as the correct document type. In unsupervised learning, the system creates the classifiers itself. Many learning systems also continue to learn in real time.

For assessing recognition quality, the simple hit rate (accuracy, classification accuracy) is not always sufficient. It is misleading in unbalanced class distributions. Unbalanced means that one type appears significantly more frequently than the others. Precision, recall, and F-score then serve as alternatives.

In SAP Business One

For e-invoices, the type of document is determined by the structured XML data. Relevant functions are the import of electronic documents and the electronic document monitor.

For paper and PDF documents, there are additional solutions such as the cks.ADC document capture module. It reads paper, PDF, and electronic invoices and generates fully completed, parked SAP documents from them. It processes incoming invoices, delivery notes, customer orders, and payment invoices. In the capture profile, the „Target Document Type“ field determines which document is created. The recommended setting „Automatic (OCR)“ detects whether an invoice or credit note is present. It does not distinguish between invoice and goods receipt. The „Base Document Type“ field describes the document that underlies the recognized document. It is relevant for linking.

In the cks.DMS document management system, categories for document types can also be preset. If a user assigns a document directly to another category in the input, the assistant’s setting is over-ridden. However, this over-setting only applies to that individual document.

Reference to Candis

A 2025 post introduces the solution Candis. According to the post, Candis classifies documents such as invoices, credits, or offers already during uploading. The software then analyzes the layout, searches for terms such as „credit“ or „notice,“ and reads out data fields such as invoice numbers and amounts. If the type is not clear, Candis, according to the post, automatically assumes „invoice.“ As a reason, the post mentions that over 80 percent of incoming documents are invoices. Users can correct the assignment, and the system, according to the post, learns from each correction. After classification, the appropriate workflow is started, according to the post.

Demarcation

Document classification determines the type of a document. OCR (Optical Character Recognition), on the other hand, recognizes the text. Data extraction reads out individual values such as invoice number or amount. According to the post, Candis also uses extracted data fields for type determination. Related entries: OCR (Optical Character Recognition), Document Information Extraction SAP Business One.

SAP B1 blog