ThoughtSpot unveils Spotter Semantics AI-native layer

ThoughtSpot has launched Spotter Semantics, a semantic layer designed for organisations deploying AI agents across analytics and business workflows. It aims to reduce inconsistent answers when users phrase natural language questions differently, or when AI systems interpret them in different ways.

Semantic layers sit between data sources and analytics tools. They provide shared definitions for metrics, dimensions, and business logic so dashboards, queries, and automated agents use the same meaning of terms such as revenue, active customer, or churn. Without that foundation, businesses can get different answers to the same question.

Spotter Semantics sits within ThoughtSpot's platform and is built around natural language search. ThoughtSpot describes it as AI-native, rather than adapted from dashboard-centric business intelligence products.

Natural language focus

The approach centres on search tokens and ThoughtSpot Modeling Language. ThoughtSpot contrasts this with systems that use large language models to convert text prompts into SQL, arguing that its method delivers more consistent results and clearer explanations of how answers were produced.

"The core challenge for modern BI agents is the lack of full context needed for precise, accurate and trusted answers. This means that organisations may be getting inconsistent insights of various qualities based not only on how questions are asked, but on how agents understand and communicate these queries. That's why a robust semantic layer has always been part of ThoughtSpot's DNA," said Francois Lopitaux, SVP of Product Management at ThoughtSpot.

"From day one we've placed an emphasis on an AI-native semantic layer that serves as the bridge between complex data and business-ready answers. Critically, this deterministic approach relies on our patented search tokens, not text-to-SQL powered by LLMs, which is why we can guarantee the most consistent, trustworthy insights on the market," Lopitaux said.

Key components

Spotter Semantics includes a query-generation engine and indexing that translate natural language requests into SQL queries. ThoughtSpot also describes a context-aware semantic architecture that references knowledge graphs. The system incorporates business logic, security rules, metric definitions, and model instructions in a machine-readable form for agents.

ThoughtSpot also highlighted support for more complex analytical structures, including multiple fact tables, formulas, cohorts, formatting, geomapping, localisation, and synonyms. It links these capabilities to fewer misinterpretations of questions by AI systems.

Governance and metrics

Governance is a central theme of the launch. ThoughtSpot said it uses a governed Metrics Catalog as a single version of truth to avoid "metric drift", where teams create similar measures with different definitions and then lose confidence in outputs.

For day-to-day administration, analysts can create custom metrics, cohorts, calendars, and formulas through a visual interface. Data engineers prepare underlying data in SQL, while developers manage deployment through APIs. ThoughtSpot also supports automated deployments with continuous integration and ThoughtSpot Modeling Language, with rollback as part of the deployment flow.

Routing and cost

Another feature is "aggregate awareness". ThoughtSpot said the system can route queries to either detail-level tables or pre-aggregated tables depending on the question, which it ties to faster responses and lower compute costs.

The release also extends what ThoughtSpot calls "next-gen search tokens". ThoughtSpot said the upgrades let its agent interpret and answer more complex questions, especially those that rely on nuanced business wording.

Open standards

ThoughtSpot said Spotter Semantics aligns with the Open Semantic Interchange initiative and is intended to remain interoperable across data and AI tooling. It positions this as a way for organisations to avoid being locked into a single vendor's semantic model.

The semantic layer can integrate with models in Snowflake, Databricks, and dbt. ThoughtSpot also pointed to the ThoughtSpot MCP server, which it said connects a governed semantic layer to AI agents and large language models through the Model Context Protocol.

Customer context

ThoughtSpot cited adoption and usage metrics for its platform and for Spotter, which it describes as an AI analyst. It said usage has increased and that most customers use Spotter as a primary analytics interface.

It also referenced comments from Sephora executive Manbir Paul, made on The Data Chief, about the role of self-service analytics and semantic layers in understanding the business concepts behind data.

"Looking at ThoughtSpot as a BI enablement for our data consumers has been a transformation journey for us, not just because of the enablement from a BI self-service perspective, but being able to get to our clients when they are exploring and trying to understand the data, and being able to enable them to capture that understanding of data. That helps us enrich our semantic layers based on how they look at data. That has driven a lot of value for us in understanding the business concepts behind the data that we have."

ThoughtSpot said its roadmap for Spotter Semantics includes writeback for what it calls actionable analytics, as well as Federated AI Search.