Create AI columns (Beta)

Create AI columns to enrich your data using natural language prompts. AI columns allow you to construct dynamic prompts that reference specific table columns, preview the resulting JSON or text output on a subset of your data, and then apply the prompt to the entire table.

AI columns can perform tasks like:

• Enrich data with additional context or details
• Generate summaries of text data
• Classify data into categories
• Translate text into different languages

Depending on the connection you are creating an AI column on, AI columns use the Snowflake’s AI_COMPLETE function or the Databricks ai_query function to generate responses.

This document covers how to create AI columns, edit AI columns, and some example AI column prompts.

You can also see AI column usage statistics for your organization, as well as set usage limits for AI columns.

System and user requirements

The ability to use AI columns requires the following:

• An AI provider must be configured for your organization. See Configure an AI provider.

• You must be assigned an account type with the Create AI column permission enabled. See Account types.

• You must be the workbook owner or be granted Can edit access to the workbook.

• You must have a Snowflake or Databricks connection with write access enabled. See Connect to Snowflake and Configure write access, or Connect to Databricks and Configure write access.

• If you are using a Snowflake connection, the user associated with the Snowflake connection must have:

  • The CORTEX_USER or AI_FUNCTIONS_USER database role. See the Snowflake documentation on CORTEX_USER database role and AI_FUNCTIONS_USER database role.
  • The USE AI FUNCTIONS privilege. See the Snowflake documentation on USE AI FUNCTIONS account privilege.

• If you are using a Databricks connection, the connection must meet the following requirements:

  • The connection cannot use Databricks SQL Classic (the ai_query function is not supported by the Classic tier).
  • If connected to Databricks SQL Pro, you must enable AWS PrivateLink to use the ai_query function.
  • Databricks Runtime 15.4 LTS or above is recommended; using 15.3 or below can result in slower performance.
  • For batch interface scenarios, Databricks Runtime 15.4 LTS or above is required.
  • Your workspace must be in a supported Model Serving region.

  For more information, see the ai_query function reference in the Databricks documentation.

Limitations

• AI columns are only supported for Snowflake and Databricks connections.
• AI column prompts cannot reference another AI column.
• AI columns cannot be used with actions.
• AI columns cannot be used in input tables.
• If you swap the data source for a table with an AI column, you need to create the AI column again.
• If you materialize a workbook with an AI column that has not finished loading, the materialization does not fail. It returns Null for the AI column rows that have not finished loading.

Create an AI column

To create a new AI column:

1. From the table you want to add the AI column to, select the down arrow to open the column menu.

2. Select Add column via > AI column….

3. If you have any filters, grouping, or sorting on your table, you are prompted to select one of the following options to add your AI column to:

   • A new child table based on the current table: The AI column is added to a new child element that is based on the current table. The child table’s data is a subset of the parent table, and only contains rows that were not excluded from the parent table (e.g. through filters). Selecting this option can improve query speed and reduce compute costs in the data platform.
   • This table: The AI column is added to the currently selected table, and generates output for all rows in the source table, including rows that may be filtered out.

After selecting an option, select Next.

4. Configure your column in the Add AI column modal:

   1. Column name: Enter a name for your AI column.
   2. Prompt: Enter your prompt. To reference an existing column in the table, press the = key on your keyboard, then search for and select the column you want to reference. You must reference at least one column in your prompt. See Example AI column prompts.
   3. (Optional) To change the model used to generate the AI column, select a model from the dropdown.
   4. (Optional) Select the Refine button to add more detail to your prompt using AI.
   5. (Optional) To format the AI column response as JSON, select the JSON format tab, then turn on the Format as JSON toggle. Enter the following information for each field you want to include in the JSON response:
      • Name: Enter a name for the field.
      • Type: Select a data type from the dropdown.
      • Description: Provide additional context for the LLM on the field’s contents and how to format the response. At least one field must be added if Response format is enabled.
   6. (Optional) From the Prompt tab, select Preview to preview the output of the first 100 rows of your AI column.
   7. (Optional) If you want the AI column to only run when manually prompted instead of automatically, select the Don’t run automatically option.
   Without the Don’t run automatically option selected, the AI column query runs automatically once you create the column, and each time the workbook is opened or refreshed. The output of the AI column is cached, so the query is only re-run for new or changed rows. Unchanged rows do not have their output regenerated.

5. Select Create. The AI column appears in your workbook, either as part of the existing table, or as part of a new child element.

See Manually run an AI column query if you selected Don’t run automatically when creating the AI column, or if you want to re-run the AI column query for a specific table cell.

Manually run an AI column query

If you selected the Don’t run automatically option when creating an AI column, you can manually run the AI column query at any time:

1. Right-click on the AI column to open to column menu.
2. From the column menu, then select Run AI column.

You can also re-run the AI column query for a single cell at any time, even if you did not select the Don’t run automatically option when creating the AI column:

1. Right-click on the AI column to open to column menu.
2. Select Re-run AI for this cell.

Edit an AI column

To edit an existing AI column:

1. From the AI column, select the down arrow to open the column menu.
2. Select Edit AI column….
3. Make your changes to the AI column configuration.
4. Select Update.

Example AI column prompts

For example, given a large table of sales representative call logs with information about the call participants, duration, and transcripts, you can create an AI column to:

• Summarize lengthy transcripts

  • Example prompt: “Write a brief executive summary of each call, with the following information included in the first sentence: the [Account Name], [Date] and [Call Duration]. Use the [Full Transcript] to summarize what went well, and any potential risks or blockers, in 2-3 sentences.”

• Classify calls into categories

  • Example prompt: “Classify the primary conversation topic for each row’s [Full Transcript] into one of the following categories: Discovery, Demo, Negotiation, Renewal, Other. Return only the category name.”

• Extract information from unstructured text

  • Example prompt: “Extract the any competitor company names mentioned by [External Participants] in the [External Speaker Transcript] and provide a 1-sentence summary for why they were mentioned.”

Set AI column token usage limits

Sigma admins can set AI column usage token limits for individual connections. If no token limit is set, the default limit is 10 million tokens per connection per month. See Configure token usage limits for AI columns for instructions on how to set token limits.

View AI column usage information

Sigma admins can view AI column usage information in Usage dashboards. AI column usage information is only supported for Snowflake connections. See Usage overview for more information.