BETA FEATURE: This is a beta feature introduced in v7.2.1. It can change as the feature evolves.
The Talk to Your Data feature enables you to submit natural language queries directly within Tableau and Power BI reports, letting you explore the full depth of the underlying semantic model without writing queries or needing analyst support. Please review this article for a comprehensive overview of how Talk to Your Data can benefit your organization, as well as guidance on preparing for its implementation.
What Is "Talk to Your Data"
Talk to Your Data is one of several Conversation Analysis features that Metric Insights provides to users. This feature uses reliable information from existing dashboards and lets users to ask Concierge questions about the data displayed in Viewers.
Concierge understands which Tableau or Power BI Element the user is viewing and answers questions based on the underlying data from those sources. Users receive responses that may include answers, charts, and trend lines generated directly from the Report’s Semantic Model.
This enables users to go beyond standard filters and slicers, supporting more advanced queries such as comparisons, anomaly detection, and metric breakdowns across dimensions. Responses appear inline next to the report, making it easy to explore follow-up questions in context.
How It Works
When viewing the External Report Viewer or Preview, users may open Concierge to submit questions regarding the displayed data. When a user submits a question, Concierge constructs VizQL or DAX queries on-the-fly against the underlying Tableau or Power BI semantic model, with no platform agents required. This eliminates the need to manually interact with a visualization by setting filters, adjusting slicers, and digging through data to surface an answer. Instead, users get results conversationally, as if asking a data analyst directly.
Concierge analyzes the retrieved data, performs comparative analysis, and delivers a detailed response to the user's inquiry. At the conclusion of the analysis, a bottom-line summary is provided, enabling users to obtain a concise answer as well.
Required Administrator Roles
| Role | Responsibilities |
|---|---|
| Tableau Administrator |
More details on Tableau configuration here. |
| Metric Insights Administrator |
|
| Role | Responsibilities |
|---|---|
| PowerBI Administrator |
More details on Tableau configuration here. |
| Metric Insights Administrator |
|
Prerequisites
Before implementing the feature, you need to consider:
- Talk to Your Data is a part of Concierge Conversation Analytic feature, which requires a Large Language Model (LLM) service provider in order to operate. You must first select and gain access to one of the following supported models:
- OpenAI models starting from GPT-5 family onward
- OpenAI GPT-5.5 model is a recommended model for Concierge
- Google Gemini Pro models
- Anthropic Claude Sonnet or Opus
- Other models with comparable reasoning capabilities supporting OpenAI chat completions API
- You also have the option to use a local LLM
- Concierge must be installed and configured on your MI instance
- Please note that using Concierge as a Data Source involves transmitting data to Concierge and, subsequently, to the LLM. Exercise appropriate caution when handling sensitive information
- For Tableau:
- The Talk to Your Data feature functions exclusively with Tableau instances that have VizQL enabled
- Configure Connected Apps authentication on your Tableau server. A secure authentication method using the User Account Token is required, rather than Service Account Tokens
- For Microsoft Power BI:
- The Azure App used in the Metric Insights Data Source must be configured for OAuth and granted the Dataset.Read.All and Tenant.Read.All permissions. The Tenant.Read.All permission is required to collect Power BI metadata. Metadata collection enables the LLM to understand the structure of the semantic model(s) and to extract measure and dimension metadata for generating DAX statements. Without this metadata, the LLM can't respond to questions about the semantic model or construct DAX for KPIs. For additional information, see Why Tenant.Read.All Permission Is Required