Making Talk to Data 3x More Accurate: How Snowflake Intelligence and Atlan Bring Context to Conversational Analytics

The promise — and problem — of conversational analytics #

Across every industry, every data team is chasing the same goal: a natural language interface to data. “Talk to Data” is becoming the new north star for analytics — an interface where anyone can ask a question in plain language and instantly explore metrics, dashboards, and insights.

Yet, most conversational analytics pilots struggle to produce consistent, trustworthy answers.

According to recent research, 49% of organizations say generative AI hasn’t met expectations, only 26% have the capabilities to move beyond proofs of concept, and just 1% have embedded AI into workflows at scale.

This mismatch between massive investments and real results defines what Atlan calls the AI Value Chasm — the widening gap between how fast AI systems are being built and how little value they deliver in production.

Through conversations with over a hundred data and AI leaders, three recurring blockers emerge, forming the Context Gap:

• Data without context: In a pilot, you can work with what you already know. A clean dataset, a narrow scope, a well-understood system. But in production, you’re dealing with distributed systems, undocumented sources, and orphaned pipelines. Context isn’t stored in one place; it’s scattered across diverse people and tools.
• AI doesn’t understand your business: Inside most organizations, meaning lives in silos. “Customer” means one thing in Product, another in Finance, and something else entirely in GTM dashboards. Everyone assumes their meaning is the default. That might be manageable in reporting, but AI doesn’t pause to ask what you meant. It acts on what it sees. And when meaning is fractured, the output sounds confident — and lands wrong.
• Governance that wasn’t built for AI: Analytics agents don’t just look at data. They take action, trigger workflows, and generate outputs that move downstream. Yet, most organizations still rely on static policies and manual reviews. There’s no real-time layer to enforce intent or adapt to how data is actually being used.

For conversational analytics, these gaps translate into inconsistent queries, hallucinated joins, and answers that can’t be traced or explained.

---

Does context really improve Talk-to-Data accuracy? #

To quantify how metadata and semantics affect AI query accuracy, Ravi Dawar and Manoj Shanmugasundaram from Atlan ran a controlled study using BirdBench, an open benchmark for natural-language-to-SQL generation.

They tested 145 queries across multiple domains, comparing:

• a baseline setup using schema-only prompts, and
• a context-enriched setup augmented with metadata — column descriptions, glossary terms, and lineage from Atlan.

The outcome:

• 3× improvement in SQL generation accuracy on complex or ambiguous queries.
• p < 2e-10, confirming statistical significance.

The result demonstrates that metadata acts as structured signal, improving an agent’s ability to interpret intent, map business meaning, and generate executable, accurate SQL.

---

Operationalizing context with Snowflake Intelligence + Atlan #

Snowflake Intelligence introduces an agentic AI experience built on Snowflake Cortex, enabling users to ask questions in natural language and get direct, grounded answers from both structured and unstructured data.

Atlan, Snowflake’s 2025 Data Governance Partner of the Year and launch partner for Snowflake Intelligence, complements this with the context layer required for accuracy, consistency, and trust — integrating metadata, semantics, and governance directly into the agentic workflow.

Together, they form an architecture where reasoning and context operate as one system:

1. Data context: Metadata Lakehouse and MCP Server #

Atlan’s MCP Server serves as the Context API for Snowflake. It delivers metadata — lineage, ownership, glossary definitions, and quality metrics — directly from Atlan’s Metadata Lakehouse, which is Iceberg-native and designed to read and write metadata at agentic-AI scale.

Moreover, Atlan Data Quality Studio, running natively on Snowflake compute, validates and monitors data quality inside the same environment to ensure data is fit for purpose for AI-powered analytics.

By keeping context close to Snowflake compute, every query and answer is grounded in real data relationships, verified sources, and quality signals. Agents can reference the correct tables, track provenance, and surface explanations inline.

2. Semantic context: Shared business language #

Conversational queries often fail because teams don’t share definitions. “Revenue,” “churn,” and “active user” can vary by department. Atlan’s Business Glossary and Domains standardize business meaning across the organization. Soon, these will connect to Snowflake Semantic Views, ensuring Snowflake Intelligence aligns natural-language interpretation with enterprise metric logic.

The result is consistent answers — the same definition of “customer” across dashboards, SQL, and AI-generated analysis.

3. Governance context: Trusted access at enterprise scale #

As agents start to reason and act, governance must move from static policies to dynamic enforcement. Atlan uniquely enables this in Snowflake through:

• Two-way tag synchronization keeps policies consistent across Atlan and Snowflake — authored once, enforced automatically.
• Atlan Policy Center turns static documentation into living policies connected to the data layer, ensuring data is always protected and incidents are raised in real-time.
• Atlan AI Governance Studio auto-discovers Snowflake AI models, apps, and agents, classifying them against frameworks like the EU AI Act.

This approach embeds risk management into the same fabric as discovery and reasoning — maintaining accuracy, compliance, and speed simultaneously.

---

Partnering to build conversational AI that actually works #

Conversational analytics is one of the most complex problems in enterprise AI. It demands precision across structured and unstructured data, consistent business meaning, and real-time governance.

As an Open Semantic Interchange launch partner, Atlan is investing in building the context layer with Snowflake. Atlan’s upcoming integration with Snowflake will leverage metadata from the Metadata LAkehouse to auto-generate Semantic Views in Snowflake. This will be enable data stewards and business users to power Snowflake Intelligence with semantic models that are complete, accurate, and automated.

This is the foundation for conversational analytics that actually works at enterprise scale.

---