Supplement to zapCash documents when using a language model
Use of the Language Model for zapCash
The language model is utilized via the Microsoft Azure OpenAI Service. This service provides REST endpoints within its own tenant, which are accessed during the project using an API key. The data sent to the service is not used for model training and is not visible to third parties. More information on privacy can be found here: Data, privacy, and security for Azure OpenAI Service - Azure AI services The model is used to analyze invoices or similar commercial documents and to extract features from these documents.
Requirements
- Software version: zapCash 1.2.6
- Project processed in zapCash
- Export of the candidate list is available to zapliance
- Optional additional analyses, if contractually agreed upon
RFC user authorization
- Access Rights: Dependent on the SAP document system, whether GOS or ArchiveLink is used. The RFC user must be set up according to the following article: SAP User Role Access Rights for zapAudit and zapCash (zapliance.com)
Setting up Azure OpenAI at the customer site
- Azure OpenAI Service: A cloud-based service from Microsoft Azure that enables companies to access advanced natural language processing (NLP) and machine learning technologies based on the OpenAI model.
- Setup requirements: Detailed guide is available. For GDPR compliance, it is recommended to select a Microsoft region within the EU.
- Provisioning: For example, from Central Sweden or France, to ensure GDPR compliance.
- Data transmission and storage: Data is transmitted encrypted. More details on privacy:
Data, privacy, and security for Azure OpenAI Service -Azure AI services - Model selection: Currently relevant version is GPT-4 (GPT4o). Detailed guide for setup is available.
- Security features: Access only with the correct API key, encrypted data transmission, optional IP address restrictions. More details on privacy: Data, privacy, and security for Azure OpenAI Service - Azure AI services
- Compliance guidelines: Responsible AI utilization strategy, see: Overview of Responsible AI practices for Azure OpenAI models- Azure AI services
- Document transcription: Performed with Azure Document Intelligence. Detailed guide is available.
Use of Azure Services by zapliance
- Required information: Endpoints and API keys.
- Further procedures: zapliance will process relevant invoice documents using the endpoints and provide the results to the relevant department. After the project ends, endpoints can be deactivated or the corresponding Azure resources deleted.
Downloading documents from SAP
- Procedure: Documents are downloaded using the zapattachmentconsole and provided to zapliance via a OneDrive exchange folder. The download is carried out together with zapliance to configure SAP connection data and commands.
Transmission of results
- Result transmission: The results of the language model are provided via import for zapCash and via SharePoint list in Teams for the VAT analysis.
Additional cost information
- Microsoft OpenAI costs: Downloading documents costs approximately €1,000-2,000. For 10,000 documents and 15,000 attachments, the costs are around €2,000.