Forum Discussion
Making CoPilot Work for our Organisation
The structure for data access, referred to as "the brain," is defined within the orchestration engine. Currently, the engine in Copilot Studio does not perform optimally when retrieving data from sources such as SharePoint. Interestingly, the orchestration engine in Copilot Studio Lite occasionally delivers better results in this regard.
Regarding the notion that Copilot reviews the first 20 relevant documents, it is important to clarify: AI is not designed to return a fixed number of documents that may be loosely related to a query. The purpose of using an agent with document integration is to ensure that high-quality backend data enables the AI to extract precise and relevant information. If the expectation is to retrieve 20 documents via AI, it must be stated clearly that this is not a suitable use case and will not yield effective results.
Additionally, generating branded documents is currently beyond the capabilities of AI. However, modern templates may offer a viable alternative in such scenarios.
Modern Templates on SharePoint
Hi,
let's dive into your questions:
- There is no deeper integration with your data in GPT-5; it is simply a more advanced language model with improved reasoning and response capabilities.
- Access to your SharePoint data is enabled by default and handled via the GraphAPI, which respects existing content permissions. When the first Microsoft 365 Copilot license is added to your tenant, a semantic tenant index is automatically created for SharePoint data. Once a license is assigned to a user, a personal semantic index is generated for their mailbox and OneDrive data.
- Regarding data protection: Microsoft does not use your data to train the underlying model. For customers within the EEA, Microsoft 365 Copilot complies fully with the EU Data Boundary and GDPR requirements. This is an enormous difference (by default) to ChatGPT.
- Regarding your requirement to generate bids based on historical data: this functionality must be configured within a dedicated agent. A default agent can only respond to queries based on the data provided. To generate such documents, very specific instructions tailored to your use case are required. We have developed various types of agents, such as those for tenders. With clearly defined instructions, the results can be significantly improved.