microsoft recently announced the low code tool Microsoft Copilot Study at Ignite 2023. Copilot Studio users can create standalone copilots and customize Microsoft Copilot for Microsoft 365, using AI-powered conversational capabilities for ad-hoc business use cases.
Copilot Studio is an end-to-end conversational AI platform that allows IT professionals and creators to create and customize co-pilots using natural language or a graphical interface. Copilot Studio users can design, test, and publish copilots that can then be leveraged within a Microsoft 365 context or for custom business purposes.
Standalone copilots are applications that address users’ natural language queries through a conversational interface. In addition to handling the dialogue with the user, co-pilots may need to retrieve information from authorized databases or execute actions on behalf of the user in external systems. Copilot Studio uses the same authoring canvas as the Microsoft Power Virtual Agent, which it replaces.
Creators implement conversational dialogs as a tree of nodes, each node representing an action (for example, display information to the user, ask a question, call an API, run a Power Automate flow). Part (or all) of a dialog tree can be made up of topics. Often a theme has a set of trigger phrases— phrases, keywords and questions that end users are likely to use to express the needs addressed by the topic.
Copilot Studio AI analyzes an end user’s natural language input and assigns a confidence score to each configured topic. The topic confidence score reflects how close the user’s input is to the topic’s trigger phrases.
For multiple topics with a confidence score higher than a confidence threshold (e.g., 85%), the end user may be prompted to select the appropriate topic (disambiguation mechanism). If only one topic exceeds the trust threshold, the dialog for that topic is executed immediately. Microsoft Copilot Studio can also delegate natural language understanding to the Azure AI Language Studio toolset.
Creators can use the generative capabilities of large language models within topic dialogs. Gary Pretty, senior product manager at Microsoft, proven how a potential Holland America Line customer might query an independent robot for information about a cruise (e.g., “Do I need a passport for my cruise?”). A creator would create that bot with just a few clicks by simply referencing www.hollandamerica.com as a key source of information. The robot would pass the end-user input to a generative model that would use the referenced content to answer the query (e.g., “Yes, you will need a passport for your cruise (…)”). The conversation continues and the bot keeps track of the context and history of the conversation so that the user can implicitly or explicitly reference past information.
This use case corresponds to what has been widely seen with generative models like ChatGPT. However, this time the bot’s response is based on the content being referenced. This grounding can help reduce erroneous responses from the robot.
Copilots can also provide a natural language interface to an application programming interface. The “Get Tours” topics are quite detailed where the bots ask the user if they have an existing reservation. The user then provides a reservation number. After that, the bot calls the corresponding API (via Power Automate) and displays its results. However, goal-oriented applications may require an amount of domain-specific manual work that correlates with the complexity of the goal (e.g., number of steps, conditions and branches, error handling and edge cases).
David Conger, senior product manager at Microsoft, provided at Ignite 2023 an example of complex API orchestration to achieve user objectives. Microsoft 365 Copilot can create PowerPoint presentations from a text document and then modify that document when prompted. Conger explained that to ensure correct identification of next steps, safe execution of identified actions, and error recovery, Microsoft turned to an Office Domain Specific Language (ODSL) that would be compatible with LLM. Microsoft 365 Copilot dynamically constructs a message within the token boundary with relevant information to help LLMs produce the correct ODSL program. The ODSL program is then analyzed, validated with automatic code correction, and transpiled to the native Office APIs, which are then executed.
Arguably, many enterprise use cases will be much simpler and fit a no-code approach. Generative AI coupled with no-code authoring tools create engaging demos for the simplest use cases. However, technology buyers may want to tie the licensing, setup, and no-code development costs associated with the technology to concrete, valuable use cases for their specific business.
Co-pilots can be distributed through various channels, including Microsoft Teams, a website, or even Skype. Microsoft Copilot for Microsoft 365 Additionally, you can take advantage of copilots created with Copilot Studio.
Manufacturers can also use multilingual co-pilots, which can communicate with customers in different languages while keeping all content in a single co-pilot. In many cases, such co-pilots can automatically detect the desired language based on the user’s web browser settings and respond in the same language.
Microsoft Power Virtual Agents (also known as Power VA) capabilities are fully included in Microsoft Copilot Studio. Co-pilot study integrates with Microsoft Azure OpenAI Studio, Azure Cognitive Services, Azure Bot Service, and other Microsoft conversational AI technologies. The Copilot Studio integration with Copilot for Microsoft 365 is now available in public preview.