Run the Analyst Agent

The Analyst Agent is an AI-powered analytics capability that enables users to explore data, create reports, and automate analyses through natural conversation.

It provides an entry point for creating automated Missions. Think of a Mission as a task, described in natural language, to guide the Analyst Agent through a series of subtasks.

🎓 Pigment Academy

Check out the courses available in the Learning Path for the Analyst Agent!

Before you begin

How the Analyst Agent can help you

The Analyst Agent lets you accelerate business report creation, by talking through your business need and, optionally, defining a repeatable Mission from your dialog. In particular, it:

• Performs detailed analysis at scale, based on a customizable set of instructions

• Autonomously conducts data drill-downs to identify key drivers of performance

• Generates fully editable reports, combining text and charts, that can be exported to PDF or used in your Boards.

Access rights and permissions

Members need the Access AI Agents permission to chat with the Agent and access Mission output, and the Configure AI Agents and Access AI Agents permission to create and run Missions, and access all Mission output, including previous output. For general information on permissions, see About Roles, Permissions, and Access Rights, and for specific information on Pigment AI permissions, see Set up Pigment AI.

Access rights are determined by the Member who runs the Mission. For example, if a Member has access rights only for UK data, the Mission output only includes UK data. Access rights are not re-checked when other Members select View output.

If the Mission output is exported to a text Metric, to control who sees the output you need to set additional access rights on the text Metric. These define who can read the contents of the cell receiving the output. For example, if a Member with access rights only for UK data runs a Mission and exports the output to a Metric with no access rights restrictions, even Members that do not have access to UK data can see that content in the Metric.

How to work with the Analyst Agent

There are two ways to access the Analyst:

1. Directly from a Board.

2. In AI Chats in the AI Sidebar.

Open an Analyst Agent chat directly from a Board

The Analyst Agent can analyze data directly from a Board. The AI Analysis symbol appears beside Metrics, Tables and Widgets of Number data type. Select this symbol to open an Analyst Agent chat in the AI Sidebar, for you to interrogate the data in your current context. Alternatively, select the floating AI button.

Chat to the Analyst Agent in AI Chats

1. Select AI Chats in the sidebar above the Application name, or the floating AI button.

2. Select Analyst from the tiles. If it is not there, ensure it is activated.

3. If you want to prevent the Analyst Agent from accessing sensitive data held in specific Metrics, turn off AI data access for them (requires Configure AI Agents and Configure Blocks permissions).

4. In the field showing Ask a question, chat with the Agent about the analysis you need. This dialogue can cover:

   • What data to analyze.

   • Which analysis steps to follow.

   • Examples of how the output should look (wording to use, whether to feature tables).

   • A structure, tone and style for the output (for example, concise, executive-friendly, bullet-pointed).

   • The language for the output. The instructions and output can be in any of 50 languages.

5. When you are done, tell it to create a Report. Use this keyword to have the Agent make a document from your dialogue, which you can share as a PDF.

6. To reuse your work, select the tile labeled Report in the Chat, and then select Save as a Mission.

7. You can re-run and manage Missions on the Missions tab of the AI Agents page (see below).

Launch and manage Missions

The AI Agents page lists all your saved Missions in the Missions tab, with information on their date of creation and run frequency. There are two ways to access this page:

• In your Application, select Settings (the three-dot menu in the sidebar), then select AI Agents.

• Select Agent Missions outputs above the Application name in the sidebar, then select Settings (the cog symbol).

This is where you can perform the following:

• Edit (for more information, see below)

• Delete

• Show History

• Launch Missions

Bear in mind that Missions can take 5 to 15 minutes to complete.

Edit a Mission

You can make manual changes to future runs of a Mission by selecting Edit Mission beside it in the Missions tab of the AI Agents page.

1. In the General tile, edit the Mission’s Name and Description.

2. In the Output type tile, select from the following:

   1. Generate a single output—to create a single, compiled output.

   2. Generate an output per Item of a Dimension—to generate reports at scale. Selecting this option requires you to specify the desired iteration Dimension in the field provided. An output report will be generated for each item in your iteration Dimension, up to a maximum of 60. Save these as Multi-Reports by exporting to a Text Metric. For more information, see below.

3. Edit the Instructions tile. Chatting with the Analyst Agent is the easiest way to fill out this field, but you can make edits to the Instructions manually here. For more information, see Mission instructions for the Analyst Agent.

4. Use Preview full instructions to check the output is as expected. Preview works by launching the Mission without saving the results. The Preview pane to the right lets you follow the Agent’s process and dialogs as it works. Select the second tab, Output, for a preview of the Agent’s report. If you want to keep these results, select Save & Launch.

5. In the Output export field, select + Add to export your output to a Text Metric for display on a Board. See Edit and share Agent Mission Reports for full details.  

6. When you have finished configuration, select Save. To save and run immediately, select Save & Launch.

Multi-Reports

Multi-Reports allow you to generate output from Missions at scale. Create customized Reports for each Item in a Dimension List or List Subset. You can iterate across a maximum of 60 items. By selecting Generate an output per Item of a Dimension under Output type, and outputting to a Text Metric structured on that same Dimension, you can export your output as Multi-Reports. Ensure the Agent refers to the current Item using the tag ‘@Current Item’ in the Mission instructions.

Troubleshoot Mission runs

• If a run fails, it is flagged in the AI Agents page, under that Mission’s History. Look for the red exclamation point next to the run, then do the following:

  1. Find the word Failed and the symbol ⓘ in the run’s third column.

  2. Hover your pointer over the ⓘ to bring up the error message.

  3. Use the error message to troubleshoot. Depending on the error, you can simply relaunch the Mission or update the Agent instructions and try again.

  4. If the error persists, select the three-dot menu for the run on the far right.

  5. Select Copy technical details from the window that pops up.

  6. Open a ticket with Support and include the copied technical details in your description of the issue.

• If your Mission is taking longer than usual, it may have encountered an execution issue. The process automatically stops and reset within two hours. You can relaunch the Mission at any time in parallel.

• If the Mission runs successfully but the results look wrong, contact Support with a clear description of the issue, pasting in the technical details and adding screenshots if relevant.

• Ensure that your Metrics have the correct default time aggregator setting. Otherwise the Agent may sum together non-cumulative data points such as FTEs or exchange rates.

• Most of the time, you will want the Agent to analyze quantitative data. But remember that you can also reference Text Metrics in the Object picker, which enrich your analyses with qualitative information. For example:

  • have the Agent take into account business comments, contextual explanations, or user-provided insights.

  • ask the Agent to quote text-based inputs verbatim in its reports alongside the quantitative results.

• Avoid referencing pre-calculated variance Metrics. The Agent computes variances itself, and using pre-aggregated values can lead to incorrect results.