Manage Spotter conversational search

Learn about Explore's optional AI-powered search experience, how to enable it, and how to improve its accuracy with coaching.

Important: Spotter is disabled by default for all users. You can choose which groups are allowed to use it. You can also enable or disable Spotter for individual data sources.

Spotter is an optional AI-powered conversational tool. With Spotter enabled, you can ask business questions using natural language and get AI-generated Answers. Then, continue the conversation with follow-up questions. If the AI Answer doesn't look right, you can edit it. If you like the AI Answer, you can add it to your Explore environment by saving it or pinning it to a Liveboard.

Spotter is located:

As a search bar on the Insights home screen, so that you can search any Spotter-enabled data source.
As a button in the top right corner of Answers on Liveboards, so that you can ask follow-up questions about an Answer. The Answer must use Spotter-enabled data sources.

Watch a user search with Spotter in this video:

GPT and data security

Spotter uses GPT by Open AI, with the large language models (LLM) of Microsoft Azure OpenAI Service.

Explore's pre-configured content doesn't include personally identifiable information (PII) or sensitive data, but if you create custom Data Studio datasets that do include it, be aware of the following:

Explore shares column metadata and sample values with GPT, including PII and sensitive data columns.
GPT doesn't store the data or use it for retraining the model. These capabilities are explicitly disabled.

To learn more about how Spotter works and its security features, see the ThoughtSpot documentation: How our natural language search works

Allow groups to use Spotter

Before you begin

You must have the Group Administration role. To learn more about groups, see Manage privileges and access.

Procedure

In the top navigation bar, use the workspace selector to go to the Admin workspace, then select Groups.
Optional: If you want to carefully limit which users can access Spotter, create a new group just for Spotter access. Then you can manage it separately from other roles.
To allow groups to use Spotter, add the Spotter Access role.

Enable or disable Spotter for individual models or tables

Before you begin

To check which data sources have Spotter enabled, go to the Insights workspace Home page. Under Spotter, look at the sources drop-down list. Any data source in this list has Spotter enabled.

To do the following steps, you must have the Spotter Access and Data Model Management roles.

Procedure

In the top navigation bar, use the workspace selector to go to the Data workspace, then select the name of a model or table to open it.
Select More > Enable Spotter or Disable Spotter.
For optimization options, select Edit Model > Spotter optimization
To learn more about optimization guidelines, see the ThoughtSpot documentation: Spotter model readiness

Allow users to manage Spotter coaching

Before you begin

To grant this permission, you must have the Data model management and Spotter access roles. The users getting the permission must already have the Spotter access role.

About this task

Users with the Data model management role already have permission to manage Spotter coaching for all models. They can give this permission to other users for individual models, without allowing them to edit the models. For example, you might want power users to manage coaching for models that they frequently use.

Procedure

In the top navigation bar, use the workspace selector to go to the Data workspace, then select the name of a model or table to open it.
Select More > Spotter coaching access.
Add users or groups, then select Save.

Coach Spotter

Before you begin

To manage Spotter coaching, you must have the Spotter Access role and one of the following permissions:

To manage Spotter coaching for all models, users must also have the Data Model Management role.
If you want users to manage Spotter coaching, but not edit models, give them Spotter coaching access for individual models.

About this task

You can coach Spotter to improve its AI-generated Answers. Coaching is specific to a data source (a model), so it's only applied when a user asks a question for that data source.

There are multiple types of coaching:

Reference questions: You ask a question, get an Answer, and ensure the Answer is correct. Then, when a user asks the same question, Spotter gives them the exact answer. Spotter also generalizes what it learns in order to improve similar questions. For example, the reference question How many claims are open in Ohio? will also help improve How many claims are closed in Maine? We recommend coaching Spotter with the most commonly asked questions.; Reference questions indirectly help Spotter learn which columns to use for answering questions, especially when there are similar column names.
Business terms: You map business terms to search tokens. Then, when a user asks a question, Spotter doesn't have to guess what their words mean; It knows exactly which data they're asking for. It's best to add terms that are specific to your organization and aren't common knowledge, because Spotter is already trained on public information. The term must also have a consistent meaning because Spotter applies the mapping to all questions asked for that model.

Coach reference questions

About this task

Spotter coaching managers can add reference question coaching at any time from the Spotter coaching page.

Procedure

In the top navigation bar, use the workspace selector to go to the Data workspace, then select Reference questions > Add reference question.
Select a data source from the drop-down list. You can Preview data to understand more about the source.
Type a question, then select Next.
Review the AI-generated Answer and edit the search tokens, chart, or table until it looks correct.
Select Add context to add a natural language explanation for your changes. Reference specific column names whenever possible.
Context helps Spotter apply the same logic to future Answers. For example, if Spotter identified assigned users with an ID, but you always want to include first and last names, add this context: When I ask for 'assigned users,' always include the Activity Assigned User column in the answer.
Select Submit Answer to create a reference question.
Optional: In the next step, Spotter predicts which business terms map to search tokens. Edit them if needed, then approve or deny them. To skip this step, select Done.
The reference question and business terms appear on the Reference questions and Business terms pages. To apply the coaching for all users, select More > Change access > Global > Apply.

Coach business terms

About this task

Spotter coaching managers can add business term coaching at any time from the Spotter coaching page.

Procedure

In the top navigation bar, use the workspace selector to go to the Data workspace, then select Business terms > Add business term.
Enter a business term that's familiar to users.
Select the data source for which you want to define the business term.
In the query search bar, enter search tokens that define the business term.
Select Submit.
The business terms appears on the Business terms page. To apply the coaching for all users, select More > Change access > Global > Apply.

Migrate Spotter coaching to another model

About this task

You can export a Spotter coaching from one model and import it into another similar model that you have access to edit.

Procedure

Use the workspace selector to go to the Data workspace, then select Reference questions.
Select the check box next to the questions or terms you want to export.
Select Export.
The reference questions TML file downloads to your computer.
Repeat steps 2-3 on the Business terms page.
The business terms TML file downloads to your computer.
Select Data objects.
Select the name of a model that you want to apply the coaching to.
Select More > TML > Import coaching TML.
Upload the TML files.
Edit the file if needed, to adjust for column or naming differences.
To learn more about resolving errors, see the ThoughtSpot documentation: Import and export coaching TML
Select Import.