FAQ for Copilot AI summarization in Commerce headquarters
This article provides answers to frequently asked questions about the Microsoft Copilot AI technology that is used to generate summaries in Dynamics 365 Commerce headquarters. It includes key considerations and details about how the AI is used, how it was tested and evaluated, and any specific limitations.
What is AI summarization in Commerce headquarters?
AI summarization in Commerce headquarters helps business users make better decisions more efficiently by giving them access to summarized information about data and risks.
What can AI summarization in Commerce headquarters do?
The feature is powered by Azure Open AI Service's large language model. It uses the information from posted and unposted statements to generate a summary. This summary includes insights such as the number of affected transactions, the total sales amount of those statements, and risks such as transactions that have returns without receipts, expense transactions, and price overrides. The AI summary is also shown for merchandisers, to provide insights about product misconfigurations. Copilot uses the gpt-3.5-turbo generative AI model to generate natural language content.
What is the intended use of AI summarization in Commerce headquarters?
The intended use is for users to have immediate access to a relevant summary of an entity, so that they can make faster, better decisions.
How was AI summarization in Commerce headquarters evaluated? What metrics are used to measure performance?
The AI-generated summary feature was manually evaluated by comparing the results against the values in current and associated entities. Further validation was done for scenarios where users are missing privileges to view data (to prevent overexposure) and scenarios where entities are missing data (to prevent hallucination). All calculations are done by using traditional coding methods, and only pre-aggregated data is sent to the large language models (LLM) service Copilot application programming interface (CAPI). The calculations were verified against the same data during the testing process. If you encounter inappropriate generated content, you can report it to Microsoft by using the Report abuse feedback form. Your feedback helps improve the functionality.
Microsoft might disable Copilot-driven features for selected customers if abuse of the functionality is detected.
What are the limitations of AI summarization in Commerce headquarters? How can users minimize the impact of those limitations when they use the system?
For very large documents, only a subset of the most impactful datasets is prefiltered, to prevent token size limitation. The generated content should never be used without manual review or supervision.
What operational factors and settings allow for effective and responsible use of AI summarization in Commerce headquarters?
When you use the feature, you should follow these recommendations:
- Confirm that your company has enough control over data permissions to ensure that business data can't be manipulated to influence AI data processing in an undesirable way.
- In the event of large priority issues, the summarization capability can be turned off from the Feature management workspace in headquarters. There are also granular controls that let you turn off summarization for individual pages, based on business need and relevance.