Spotter, ThoughtSpot’s AI-powered analyst, has redefined how users interact with data by answering natural language queries in seconds. While Spotter works efficiently right out of the box, taking the time to coach it can significantly enhance its accuracy. For organisations with complex data structures or unique business definitions, this step ensures Spotter delivers highly relevant, precise, and actionable insights.
In this guide, we’ll explore why coaching Spotter is so important, the benefits it offers, and how to fine-tune it to align perfectly with your organisation’s needs.
Why Coach Spotter?
While Spotter is capable of connecting to data and providing insights immediately, coaching it takes things a step further. It ensures Spotter understands the specific logic, filters, and definitions unique to the organization’s datasets. This means more accurate answers, reduced manual effort, and smoother processes.
Here are the key reasons why coaching Spotter is worth the investment:
- Improved Data Accuracy: Coaching helps Spotter learn the specific nuances of an organization’s data. For example, if "Revenue" always requires a particular filter to exclude refunds or discounts, Spotter can be coached to apply that logic automatically. This eliminates inconsistencies and ensures users always get the correct results.
- Simplified User Experience: Without training, users may need to have a deep understanding of the data structure to phrase their queries correctly. Training Spotter removes this burden, allowing even non-technical users to get accurate insights without worrying about the underlying data intricacies.
- Streamlined Workflows: Spotter reduces the back-and-forth with analysts by providing accurate answers upfront. This not only saves time but also empowers users to access insights independently, freeing up analysts to focus on more strategic tasks.
Now that the benefits are clear, let’s move on to the actual training process and how it can be implemented effectively.
The Coaching Process: Step-by-Step
Coaching Spotter involves refining how it interprets questions and connects them to the organisation’s data. This includes setting up reference questions, defining key business terms, and adjusting tokens to ensure Spotter understands queries the way the organisation intends.
Pre-requisite – Preparing the data
To enable Spotter to work with your data Model in ThoughtSpot, you'll first need to activate it. This is a simple step—just navigate to the Model, open the menu, and select Enable Spotter.
Once Spotter is enabled, it's important to ensure that the relevant data is properly set up. To help Spotter accurately interpret users' questions and terminology—and to deliver the best possible experience—you should add meaningful field descriptions and synonyms before starting the coaching process. This setup is done at the Model level within the Data Workspace. Simply select the appropriate Model, then manually add descriptions and synonyms for each field.
This step is key: users may refer to a single field using different terms, and Spotter needs to recognize they're all referencing the same thing. Likewise, clear and thorough field descriptions help Spotter understand the context and purpose of each field, resulting in more accurate and useful responses.
While not strictly required, it's strongly recommended to index attribute columns in both Worksheets and Models to improve search accuracy. Proper indexing allows ThoughtSpot to retrieve and display sample values for each column. These sample values, along with column names, are sent to the LLM, helping it:
- Better understand each column's properties
- Accurately match user queries to the correct column values when generating responses
To do this, simply click the Optimise for Spotter button. This makes it a lot easier because it tells you columns which are not configured to index values and allows you to easily configure indexing on the appropriate columns.
Step 1: Establishing Reference Questions
With the data now AI ready, the first step in coaching Spotter is creating reference questions. These serve as examples for Spotter to learn how to interpret and respond to queries correctly.
In the Data Workspace, navigate to the Spotter Coaching section and then the “Reference Questions” page. Here, users can input sample queries that represent common questions asked by their team, such as “How many jackets were sold last year?”
When Spotter receives a question, it breaks it down into components known as tokens. For example, in the query above:
- Filters: Spotter identifies “last year” as a date filter and “jackets” as a product filter.
- Measure: Spotter recognizes “How many” as a request for the sum of the quantity sold.
By validating and refining these tokens, Spotter learns how to handle similar questions in the future. For instance, once coached, it will consistently interpret “What’s the total jacket sales from last year?” in the same way.
This step is critical because it builds the foundation for Spotter’s ability to deliver accurate results.
Step 2: Defining Business Terms
The next step is to define key business terms within the platform. This ensures Spotter understands the specific meaning of phrases and words used in the organisation.
For example, consider the phrase “sold.” In one organisation, it might refer to the sum of quantity purchased, while in another, it could mean net sales revenue. In the “Spotter Business Terms” section, these definitions can be set so Spotter automatically applies the correct logic.
Similarly, common phrases like “last year” or “this quarter” can be mapped to specific date filters, ensuring Spotter applies these filters accurately without requiring additional user input.
By defining these terms upfront, Spotter eliminates ambiguity and ensures consistency across all queries.
Step 3: Refining Tokens
Sometimes, Spotter may need further adjustments to handle more complex data structures. This is where token refinement comes in.
For instance, if a dataset contains both “order date” and “delivery date” fields, Spotter might need guidance on which one to prioritize based on the context of the query. Users can refine these tokens within Spotter by specifying preferences or resolving conflicts.
Refinement also allows adjustments to handle edge cases, such as when a dataset has overlapping field names or ambiguous relationships. By resolving these issues during coaching, Spotter becomes much more reliable when handling real-world queries.
Step 4: Testing and Saving
Once the reference questions are established, business terms are defined, and tokens are refined, the final step is to test the changes and save them.
Testing involves running sample queries to ensure Spotter responds as expected. For example, after coaching, running a query like “What’s the revenue for jackets last year?” should return a precise result, automatically applying all the defined logic, filters, and measures.
If any discrepancies arise, further adjustments can be made to fine-tune Spotter’s understanding. Once everything works as intended, the updates are saved, and Spotter begins applying the refined definitions to all future queries.
Step 5: Monitoring Spotter Conversations
If you want to monitor the current Spotter usage, you can access the “Spotter Conversations” Liveboard.
Here, you can see how the general accuracy of Spotter is, what kind of Questions are right and wrong, and which users are using Spotter frequently.
Make the most of ThoughtSpot’s features to uncover deep, granular insights. You can explore key areas to discover where your agents are truly adding value!
Summary
Coaching Spotter is more than a technical task—it’s about making the AI smarter and more aligned with the organization’s needs. By establishing clear reference questions, defining important business terms, and refining tokens, Spotter becomes a tool that provides consistent, precise, and actionable insights.
This not only simplifies the user experience but also empowers teams to explore data independently, driving better decisions across the organisation. With a well-coached Spotter, organisations can ensure their data works for them - not the other way around.
Need help training Spotter in ThoughtSpot?
At 7Dxperts, we specialise in uncovering the full potential of ThoughtSpot and the modern data stack. Our expertise spans across the most advanced data technologies, including:
- Accelerators for ThoughtSpot: Check out our Dbx2Spot and Snow2Spot solutions to fast-track your ThoughtSpot implementation and make Spotter more accurate .
- Geospatial Analytics: Unlock insights with geospatial data using our Map2Spot accelerator, designed for platforms like #Databricks, #Snowflake, #BigQueryand #Redshift.
- Data Readiness Expertise: We will support you through modernisation, augment with your organisation with a view to handover and partner as advisors over a longer period of time.
Whether you’re just getting started with ThoughtSpot or looking to optimise your analytics strategy, 7Dxperts is here to help you achieve your goals faster and more efficiently.
