Extensibility

Templates for Document Information Extraction

Document Templates and AutoML with SAP AI Business Services

SAP AI Business Services can be consumed in enterprise landscapes using SAP Business Technology Platform (SAP BTP) offerings, such as SAP Cloud Integration, SAP Workflow Management, or Business Rules. For example, Data Attribute Recommendation and Document Information Extraction integration packages are available in Cloud Integration. Additionally, combining the services with SAP Intelligent RPA enables the full automation of end-to-end business processes.

The Intelligent Invoice Scanning feature from SAP Business ByDesign is using the Document Information Extraction from SAP AI Business Services with an overall accuracy reaching 87.75% as per Q1 2021 statistics. This accuracy rate is based on the usage of over 100 productive customers which adopted the new feature in the last 7 months since GA release, which are scanning almost 35,000 invoices per month as per March 2021. In addition, the number of perfectly scanned invoices (100% header accuracy) reached 43%, while almost 80% of the scanned invoices have a header accuracy of over 90%.

Simona Marincei, Head of Intelligent Processes, Product Management SME

SAP is releasing new features for the Data Attribute Recommendation and Document Information Extraction business services for SAP AI Business Services:

  • The Document Information Extraction service can extract business-relevant entities (header fields and line items) from documents such as invoices, payment advices, and purchase orders. This helps organizations reduce manual and error-prone work and automate business document processing. With its new templating functionality, users can create custom templates based on the defined layout structure of a business document. The functionality accelerates the processing of high-volume document types from business partners that do not change their document layouts often. The templating leads to higher extraction accuracy, drives efficiency and effectiveness, and simplifies overall business document processing efforts. This feature is especially relevant for organizations and units processing high amounts of documents (for example, shared service centers, which usually process documents that arrive in the same format from different customers).
  • The Data Attribute Recommendation service saves manual effort and accelerates processes by automatically predicting relevant fields or decisions – for example, within material master data or transactional data (such as prediction of material classes for newly created materials), or workflow decisions (such as approvals or rejections). The service now offers a new model template that uses automated machine learning (AutoML) approaches for predictions to automatically pick the machine learning model and configuration with the best-possible prediction accuracy. The AutoML model template supplements existing model templates provided and expands the set of use cases the service can handle. This is done without the need for an expert data scientist.