Microsoft Fabric Copilot is a cutting-edge AI-powered tool integrated with Microsoft Fabric and Power BI, designed to revolutionize the way businesses interact with and analyze their data. Leveraging Azure OpenAI services, Microsoft Copilot introduces a range of generative AI capabilities that allow users to effortlessly manipulate, explore, and visualize data through natural language inputs. Whether creating detailed reports in Power BI or transforming data in Data Factory, Copilot enhances productivity by automating complex tasks and delivering insightful outputs. This tool is being rolled out progressively to customers with paid Fabric or Power BI Premium capacities, offering powerful AI-driven features while maintaining strict data privacy and security standards. Although it provides significant advancements, it is essential for users to carefully review Copilot's outputs to ensure accuracy and reliability in their work.
Enable Microsoft Fabric Copilot
The Microsoft Copilot and other generative AI features currently in preview offer unique opportunities to manipulate and examine data, produce conclusions, and develop visual representations and reports within Microsoft Fabric and Power BI.
Microsoft Copilot capabilities in Microsoft Fabric are enabled by default in the Fabric admin portal. However, if your organization isn't ready to adopt Copilot, it can be disabled.
Microsoft Copilot in Microsoft Fabric is rolling out in stages with the goal that all customers with a paid Fabric capacity (F64 or higher) or Power BI Premium capacity (P1 or higher) have access to Copilot. It becomes available to you automatically as a new setting in the Fabric admin portal when it's rolled out to your tenant. When charging begins for the Copilot in Fabric experiences, you can count Copilot usage against your existing Fabric or Power BI Premium capacity.
Business data
Copilot features use Azure OpenAI Service, which Microsoft fully controls. Your data isn't used to train models and isn't available to other customers.
You retain control over where your data is processed. Data processed by Microsoft Fabric Copilot stays within your tenant's geographic region unless you explicitly allow data to be processed outside your region—for example, to let your users use Copilot when Azure OpenAI isn't available in your region, or availability is limited due to high demand.
Microsoft Fabric Copilot does not store your data for abuse monitoring. To enhance privacy and trust, we’ve updated our approach to abuse monitoring: previously, we retained data from Microsoft Fabric Copilot, containing prompt inputs and outputs, for up to 30 days to check for abuse or misuse. Following customer feedback, we’ve eliminated this 30-day retention. Now, we no longer store prompt-related data, which demonstrates our unwavering commitment to your privacy and security.
Check Copilot outputs before you use them.
Copilot responses can include inaccurate or low-quality content, so make sure to review outputs before you use them in your work.
People who can meaningfully evaluate the content's accuracy and appropriateness should review the outputs.
Today, Microsoft Copilot features work best in English. Other languages may not perform as well.
How Copilot works
The term "Copilot" encompasses a variety of generative AI functionalities and capabilities within Fabric that are supported by Azure OpenAI Service. These features are primarily intended to produce natural language, code, or other content based on:
(a) the inputs provided, and
(b) the grounding data available to the feature.
For instance, Power BI, Data Factory, and data science offer Microsoft Copilot chats that allow users to ask questions and receive responses contextualized to their data. Microsoft Copilot for Power BI can generate reports and other visualizations, while Copilot for Data Factory can transform data and explain the steps it has applied. Additionally, data science offers Microsoft Copilot features beyond the chat pane, such as custom IPython magic commands in notebooks. In the future, Copilot chats may be integrated into other experiences within Fabric and other features powered by Azure OpenAI behind the scenes.
This information is sent to Azure OpenAI Service, where it's processed, and an output is generated. Therefore, data processed by Azure OpenAI can include:
The user's prompt or input.
Grounding data.
The AI response or output.
Grounding data may include a combination of dataset schema, specific data points, and other information relevant to the user's current task. Review each experience section for details on what data is accessible to Copilot features in that scenario.
Interactions with Microsoft Copilot are specific to each user. This means that Microsoft Copilot can only access data that the current user has permission to access, and its outputs are only visible to that user unless that user shares the output with others, such as sharing a generated Power BI report or generated code. Copilot doesn't use data from other users in the same tenant or other tenants.
Microsoft Fabric Copilot uses Azure OpenAI—not the publicly available OpenAI services—to process all data, including user inputs, grounding data, and Copilot outputs. Microsoft Fabric Copilot currently uses a combination of GPT models, including GPT 3.5. Microsoft hosts the OpenAI models in the Microsoft Azure environment, and the Service doesn't interact with any services by OpenAI, such as ChatGPT or the OpenAI API. Your data isn't used to train models and isn't available to other customers.
Microsoft Fabric Copilot Process
These features follow the same general process:
Microsoft Fabric Copilot receives a prompt from a user
The prompt might be presented as a query typed by a user in a chat window or as an activity, such as clicking a button labeled "Generate a report."
Microsoft Fabric Copilot preprocesses the prompt through an approach called grounding
This could involve obtaining important information, such as dataset structure or conversation history, from the user's ongoing session with Microsoft Copilot in certain situations. Grounding enhances the precision of the request, ensuring that the user receives pertinent and feasible suggestions for their particular task. Data retrieval is limited to information that the authenticated user has permission to access.
Microsoft Fabric Copilot takes Azure OpenAI's response and postprocesses it
Depending on the scenario, this postprocessing might include responsible AI checks, filtering with Azure content moderation, or additional business-specific constraints.
Microsft Fabric Copilot returns a response to the user through natural language, code, or other content
For instance, a reply could come as a chat message or created code, or it could be an apt form considering the context, like a Power BI report or a Synapse notebook cell.
The user reviews the response before using it
Copilot's responses may contain inaccurate or low-quality content, so it's crucial for subject matter experts to verify the outputs before using or sharing them.
Just like each experience in Fabric is designed for specific scenarios and user types, ranging from data engineers to data analysts, each Copilot feature in Fabric has also been developed with distinct scenarios and users in mind.
For information on the capabilities, intended uses, and limitations of each feature, please refer to the section for the specific experience you are working in.
Definitions
Prompt or input
The text or action submitted to Copilot by a user. This could be in the form of a question that a user types into a chat pane or in the form of an action such as selecting a button that says "Create a report."
Grounding
A preprocessing technique where Copilot retrieves additional data that's contextual to the user's prompt, and then sends that data along with the user's prompt to Azure OpenAI in order to generate a more relevant and actionable response.
Response or output
The content that Copilot returns to a user. For example, a response might be in the form of a chat message or generated code, or it might be contextually appropriate content such as a Power BI report or a Synapse notebook cell.
What data does Copilot use, and how is it processed?
When generating a response, Microsoft Copilot utilizes the user's prompt or input, along with additional data obtained through the grounding process. This combined information is then sent to Azure OpenAI Service for processing, resulting in an output. The data processed by Azure OpenAI may include the user's prompt or input, grounding data, and the AI response.
Grounding data encompasses a mix of dataset schema, specific data points, and other pertinent information related to the user's ongoing task. Each experience section details the accessible data for Copilot features in that specific scenario.
Interactions with Copilot are personalized to each user, ensuring that Copilot can only access data authorized by the current user. Furthermore, the outputs generated by Copilot are visible only to the respective user unless shared with others. Copilot does not utilize data from other users in the same tenant or tenants.
Microsoft Fabric Copilot leverages Azure OpenAI to process all data, comprising user inputs, grounding data, and Copilot outputs. Currently, Copilot utilizes a range of GPT models, including GPT 3.5, hosted in Microsoft's Azure environment. The service does not engage with other OpenAI services like ChatGPT or the OpenAI API. Moreover, user data is not employed to train models and is inaccessible to other customers.
Data residency and compliance
You have the authority to determine the location where your data is handled. Data that Copilot processes in Fabric is kept within the geographic region of your tenant unless you explicitly authorize the processing of data outside your region. This may be done to enable your users to use Copilot when Azure OpenAI is unavailable in your region or when availability is restricted due to high demand. To allow data to be processed elsewhere, your administrator can activate the setting that permits data sent to Azure OpenAI to be processed outside the geographic region, compliance boundary, or national cloud instance of your tenant.
How to use Copilot responsibly
Microsoft is dedicated to upholding its AI principles and Responsible AI Standard, ensuring that its AI systems empower customers while being used in a responsible and effective manner. The company's commitment to responsible AI is continuously evolving to proactively address new challenges.
The Microsoft Fabric Copilot is designed to meet the Responsible AI Standard and undergo a thorough review by diverse teams to identify and mitigate potential harms.
Copilot for Data Factory is an AI-powered toolset designed to empower novice and experienced data handlers to optimise their operations. It facilitates intelligent code generation for seamless data transformation and offers comprehensive code explanations to aid in understanding complex tasks.
Copilot for Data Warehouse
Microsoft Copilot for Synapse Data Warehouse is a powerful AI assistant built to enhance and simplify your data warehousing operations. Some of the standout features of Copilot for Warehouse comprise Natural Language to SQL, code completion, rapid actions, and intelligent insights.
Copilot for Power BI
Power BI has recently incorporated generative AI, enabling users to effortlessly generate reports by either selecting a report topic or prompting Copilot for Power BI on a specific subject. With Microsoft Copilot for Power BI, users can easily produce a summary for their report page and create synonyms to enhance Q&A capabilities.
Copilot for Real-Time Intelligence
Microsoft Copilot for Real-Time Intelligence is a cutting-edge AI tool created to assist in data exploration and extraction of valuable insights. Users can input data-related questions, which are then seamlessly converted into Kusto Query Language (KQL) queries. This tool simplifies the data analysis process for both seasoned KQL users and citizen data scientists.
Build your own copilots
Using the client advisor AI accelerator tool enables the development of personalized copilot solutions tailored to your enterprise data. The client advisor AI accelerator creates comprehensive copilot solutions by utilizing Azure OpenAI Service, Azure AI Search, and Microsoft Fabric. This all-in-one custom copilot equips client advisors to harness generative AI capabilities, optimising daily tasks and enhancing client interactions across structured and unstructured data.
Conversational knowledge mining solution accelerator
The conversational knowledge mining solution accelerator leverages Microsoft Fabric, Azure OpenAI Service, and Azure AI Speech to help customers analyze vast amounts of conversational data using generative AI. By identifying key phrases and operational metrics, this solution enables users to uncover valuable insights that can drive business impact.
How do I use Copilot responsibly?
Microsoft is dedicated to ensuring our AI systems adhere to our AI principles and Responsible AI Standards. These principles aim to empower our customers to use these systems effectively and in accordance with their intended purposes. Our approach to responsible AI is constantly developing to proactively tackle emerging issues.
The Microsoft Copilot features in Fabric are designed to comply with the Responsible AI Standard. This means that they undergo a thorough review by multidisciplinary teams to identify potential harms and are then refined to include mitigations for those harms.
Before accessing Copilot, your administrator must enable it in Fabric.
Microsoft Fabric Copilot Pricings
Started from March 1, 2024, Microsoft Fabric Copilot is billing as part of your existing Power BI Premium or Fabric Capacity. No need for separate capacity purchases—just integrate and go! Let’s break down how the pricing works and what it means for you.
How Fabric Copilot Consumption Works
Imagine having a smart assistant in your data workspace, ready to help you analyze and visualize your data effortlessly. That’s what Fabric Copilot offers. It processes your text inputs and outputs, and the cost is based on the number of tokens it handles.
What Are Tokens?
Tokens are like the building blocks of your text. They include everything from characters and words to punctuation and spaces. For instance, the sentence “Hello, world!” is broken down into four tokens: “Hello”, “,”, “world”, and “!”.
Input Tokens: This covers the text you enter into Copilot’s chat pane, including metadata like file paths and table details.
Output Tokens: These are the responses, code snippets, or Power BI reports that Copilot generates for you.
How Much Will It Cost?
Here’s the scoop on pricing:
Input Tokens: 400 capacity units per 1,000 tokens
Output Tokens: 1,200 capacity units per 1,000 tokens
Calculating Your Costs
Wondering how to estimate your costs? Use this simple formula:
Fabric Copilot consumption (CU seconds) = (input token count 400 + output token count 1,200)/ 1,000.
This formula will help you gauge the expense based on how many tokens are processed during your interaction with Fabric Copilot.
Limitations of Copilot
Copilot responses may sometimes contain inaccurate or low-quality content, so it is important to carefully review outputs before using them in your work.
It is recommended that individuals with the ability to assess the accuracy and appropriateness of the content are responsible for reviewing outputs.
Copilot's performance is currently optimized for the English language, while its performance in other languages may not be as robust.
Available Regions
Available regions for Azure OpenAI service
To use the prebuilt Azure OpenAI Service, including the Copilot in Fabric, you need a paid SKU (F64 or higher, or P1 or higher) with available capacity in specific Fabric regions. It's important to note that the Azure OpenAI Service is inaccessible using trial SKUs.
The Azure OpenAI Service utilizes large language models currently only deployed to US datacenters (East US, East US2, South Central US, and West US) and the EU datacenter (France Central). If your data is located outside of the US or EU, this feature will be disabled by default unless your tenant admin enables the "Data sent to Azure OpenAI can be processed outside your capacity's geographic region, compliance boundary, or national cloud instance" tenant setting.
Data processing across geographic areas
The Azure OpenAI Service and Copilot in Fabric may handle your prompts and results (input and output when using Copilot) in a geographic region different from your own, depending on the hosting location of the Azure OpenAI service. The table provided outlines where data is processed across geographic areas for Copilot in Fabric and Azure OpenAI features.
Conclusion
Microsoft Fabric Copilot is designed to enhance your data handling experience with powerful AI capabilities, making it easier to generate reports and analyze data. While it provides valuable assistance, reviewing its outputs carefully is important to ensure accuracy and relevance. By understanding its features and limitations, you can make the most of Copilot while keeping your data secure and compliant. As Microsoft Copilot continues to evolve, it promises to offer even more advanced tools to streamline your data processes.
Comments