What M365 Copilot is and how it works
Microsoft 365 Copilot differs from the chatbots you may already know in that large language model technology is integrated into the Microsoft 365 applications many people already use every day, such as Word, Excel, PowerPoint, Outlook and Teams. It thus represents an important trend referred to as ‘integrated AI’, i.e. AI solutions and functionality that appear in products you already use.
Sometimes this is a functionality you can choose separately (M365 Copilot requires its own additional license). Other times, you get the technology ‘thrown in’ whether you want it or not. An example of the latter is the formerly named ‘Bing chat’, which is now also called ‘Copilot’, but which is a separate chatbot solution similar to, for example, ChatGPT. This type of solution can answer general questions and does not by default have access to the user’s and the organisation’s information.
M365 Copilot, on the other hand, acts as a personal assistant and combines the functionality of large language models with an organisation’s internal data and information structures, made available via knowledge graphs and indexed semantic searches (RAG – see Appendix 1). The tool can also be configured to retrieve external information (from the Internet) to ‘enrich’ internal searches. It is worth noting that one of the potential challenges associated with this is that prompts that are initially perceived as internal or even confidential are suddenly exposed externally as searches in open search engines. M365 Copilot is a product that requires the organisation to have extremely good control of the solution’s abilities and limitations in practice.
It is worth noting that the term ‘Copilot’ is used by Microsoft in many different contexts and does not refer to a specific product. Rather, it is a support tool implemented in different ways in different parts of the Microsoft 365 platform. The promise from Microsoft is that M365 Copilot can help you with everything from generating content and analysing data to improving communication and collaboration in your organisation – in return for giving it access to all the information you have access to.
Knowledge graphs (Microsoft Graph) and semantic searches constitute a data platform that connects data and services across Microsoft 365. It collects information from emails, calendars, documents, meetings, chats and more. Through Microsoft Graph, Copilot can access and understand the context of the information a user has access to. In addition, the solution builds an experience-based profile of the user over time, which is intended to provide more relevant and personalised assistance.
Because M365 Copilot is tightly integrated into the Microsoft 365 applications, the solution is adapted to the interface and work tasks of the various tools and it can therefore assist directly in the application without having to move information between different applications or interfaces.
By automatically retrieving information from meetings, emails, documents and chats (within given access rights) in combination with user profiling over time, the solution has the potential to provide more tailored assistance. This rather extensive access to the individual user’s and the organisation’s information will make it possible to automate complex tasks:
• Outlook: Summarise long email threads. Generate email response suggestions based on context, tone, and previous communication. Identify and suggest calendar appointments or tasks based on the content of your emails.
• Word: Prepare draft documents based on prompts or key points. Improvement of text, wording, grammar and style, and adaptation of tone to the target audience. Summary of documents and reports.
• Excel: Interpret datasets, identify trends and patterns, and present layouts in a simple manner. Create complex formulas by describing what you want to achieve in natural language. Suggest appropriate charts and graphs to represent data visually.
• PowerPoint: Generate entire presentations based on a document or idea, including suggested text, images and design. Provide recommendations for layout, colours and graphics. Convert a document or report to a presentation by extracting the key points.
• Teams: Summarise discussions in real time, write minutes, and identify action items. Generate a list of next steps and assign tasks to team members.
• SharePoint: Help to create and edit content, including text suggestions and structuring. Semantic searches to find relevant content based on meaning, not just keywords. Analyse content across SharePoint to identify knowledge gaps or overlapping information.
In other words, a lot of functionality is promised, and it is all made possible by two things: the power of large language models and more or less free access to all the information the individual user in the organisation already has access to.