Copilot Studio combines the creation of targeted AI assistants with a new way to create chatbots for the Microsoft 365 environment and beyond quickly.
Copilot Studio is part of the suite of Microsoft’s Copilot applications announced last year, and it, along with other Copilot applications, has taken great strides forward this year.
In general, when talking about Copilot, we mean Microsoft 365 Copilot, which uses a large language model, the company’s data from the M365 environment, and the Internet in various applications such as the browser, email, or Teams.
Copilot Studio, on the other hand, offers the possibility of creating customized bots whose knowledge is based on data sources chosen by the user. Using Copilot Studio, creating bots is fast, and it allows testing the bot during creation.
For example, public websites, individual documents, and data from the M365 environment, such as SharePoint sites, OneDrive, and Dataverse, can be used as data sources.
In addition, Copilot Studio enables connection to other company data sources, such as SQL databases or the creation of customized connectors. These possibilities allow for a much more comprehensive data range than the M365 Copilot.
Available data sources in Copilot Studio:
- Public Website
- Documentaries
- SharePoint
- OneDrive
- Dataverse
Data sources in preview mode:
In addition to a wide selection of data sources, Copilot Studio’s game-changing feature triggers actions based on conversation. It enables, for example, the integration of Power Automate workflows into the bot. In this way, the user can optionally start automation from a conversation that, for example, sends an email, updates projects, or retrieves up-to-date information from tabular data.
Data sources and functions can be accessed through the language model or topics. Topics enable the creation of discussion dialogs or the connection of functions to them.
They can be triggered using keywords or phrases such as “Book classes” or “Find project information.” After that, conversation paths can be scripted into them; for example, in the work hours recording system, the automation that records the hours can be connected to the path with the hours given by the user as the input.
Adaptive cards are short, lightweight JSON-based user interface snippets that can be embedded in a conversation. They can query and validate information provided by the user.
In the example below, the time recording information is entered using a card so the user knows what information is being entered. Cards can be embedded with images, drop-down menus, and keys, as well as validate user-provided information such as email addresses.
The language model of a customized Copilot can be improved by utilizing and adding entities that reflect general entities or terms. By default, several of them, such as age, language, or file, describe commonly used entities.
These can be supplemented with the company’s specific terms or listings, such as product codes or categories. Entities help the bot understand the conversation better.
There are three possible ways to identify Copilot users:
- No identification
- Identification with Entra ID (only in Teams or Power Apps applications)
- Manual login (login in other applications)
It would be best if you left out the identification of whether the bot will be used by users outside the company, for example, on websites. If the application becomes available for the company’s own users, Entra ID is the easiest way as it recognizes the user’s information automatically. If the application will be used elsewhere and you want to identify users, choose option 3.
Copilot Studio currently supports the Finnish language, but some functions are only available in English.
Janne Häkkinen
Business Intelligence Consultant
