Extending Microsoft 365 Copilot: Plugin Options & Future Possibilities

Would you like Microsoft 365 (M365) Copilot to provide more features and information for your organization’s needs? Can we make the M365 rule for all business needs and use cases?

A few days ago, I had an opportunity to discuss and present some of my learnings and understanding of extending the M365 Copilot at the CollabDays Finland event in a breakout session about ‘Copilot extensions & their possibilities.’

The original post can't be found on my blog: Extending Microsoft 365 Copilot: Plugin Options & Future Possibilities – Mikko Koskinen

Different Extension Options

As we know, the Microsoft 365 Copilot (M365 Copilot) is a capable tool with its limitations. And I don’t expect it to do everything we users hope with out-of-the-box features. Microsoft offers tools to tweak and extend the function based on the organization’s needs.

From a technical point of view, there are several options to extend the capabilities of Microsoft 365 Copilot. The main categories are extending the semantic index and knowledge of the Copilot; the other is adding new skills to the bot through the plugins.

There is a fundamental difference between these two models.

• The plugins are something that the users can decide to use by activating or de-activating them in the Copilot Pugin section. Copilot will choose when to use the active plugin based on the intent of the conversation.
• Graph connector data, on the other hand, is always available, and the Copilot will use the information on the answer whenever it sees it valuable in the context of the conversation.

Also, at the time of writing this post, the Plugins are only available in the Teams version of M365 Copilot.

Teams Message Extensions: These powerful Microsoft Teams and Outlook features allow users to interact with web services directly from the chat interface.

With Copilot for Microsoft 365, natural language input can invoke a message extension’s search function, bypassing the need for specific UI commands. Users can search or initiate actions in external systems from the compose message area or directly from a message.

The message extension must be built using Visual Studio and Pro development knowledge. More information: Build message extensions for Microsoft Copilot for Microsoft 365 | Microsoft Learn

Graph Connector: This feature grounds Copilot interactions in extended data by indexing content from external systems into Graph and the Semantic Index. This allows Copilot to provide more complete and contextualized responses based on data that is even outside of the Office 365 context.

The data brought through the Graph connector will be available for users in Copilot and Microsoft Search service. There are multiple ready-made Graph connectors available from different ISV systems, and you can also create your custom connector with pro-coding methods: Build Microsoft Graph connectors for Microsoft Copilot for Microsoft 365 | Microsoft Learn

Copilot Studio Plugins: By leveraging Microsoft Power Platform capabilities, these plugins extend Copilot for Microsoft 365’s skills. Users can create conversational actions, run Power Automate flows as plugins, and use custom prompts to instruct the GPT model to perform specific tasks.

Unlike other plugin options, Copilot Studio plugins are also available for citizen developers because they leverage the Power Platform’s low-code capabilities. If you have ever created a Power Automate flow, you are ready to extend the M365 Copilot: Extend Copilot for Microsoft 365 with copilot extensions – Microsoft Copilot Studio | Microsoft Learn

Upcoming Extensions: And the development won’t stop here. Other extension possibilities are coming that are currently under a private preview. API plugins enable Copilot for Microsoft 365 to interact with REST APIs described by an OpenAPI specification. Declarative Copilot allows you to craft your own by declaring instructions, actions, and knowledge running on the same orchestrator that powers Microsoft Copilot.

Thoughts and Learnings

One important part of extension building is the description and manifest for the plugin. Based on this information, the Copilot decided when to use the plugin. I’ve found that there is still considerable variance in when the plugin is used. Although the users think they are promoting in a way that would give them information through the plugin, it might be that the Copilot believes the other way.

I hope Microsoft will offer us more tools and control over when the plugins are used.

We must also remember that we are still in the early days of these extensions. We are constantly learning more, and Microsoft is continually building new features. After discussing with a tenth of organizations and facilitating multiple workshops, I can say that the need and momentum exist.

• Be open-minded and allow the people in your organization to innovate new things and tell about their needs. It’s incredible what kind of services I’ve seen people create while you give them the opportunity and time to do so.
• Remember, not everything needs an AI chatbot for usefulness and benefit – the “old” ways are still valid. In many cases that have started as an AI discussion, I’ve ended up recommending something else like automation or process building.
• Don’t be afraid to build your own Copilot tools and AI processes.M365 Copilot is an excellent tool, but there is always a time and place for a standalone tool, for example, when you want to be sure that only certain information is used as a data source.
• And finally, there are no shortcuts, even in AI – if there is no API or connecting to some data, AI can’t operate without it.Also, the quality of your data in general is still essential, which could mean that you first need to work in a data consolidation project before AI can help you efficiently.