Snowflake CoCo Defined
Snowflake CoCo is the conversational AI assistant Snowflake announced at Summit 2026. It lets a business user type a natural-language question like "what was Q3 gross margin for North America?" and get an answer that comes from inside the Snowflake data cloud, governed by Horizon Context, and grounded in the metric definitions the data team has approved.
A conversational AI assistant for Snowflake data, integrated with Horizon Context for governed business meaning.
A standalone chatbot. CoCo's answers are only as good as the semantic definitions and governance underneath it, so the same product can produce trusted answers in one environment and unreliable ones in another depending on how its context layer is set up.
Analysts, business users, and the AI agents that need a governed query layer instead of a raw SQL connection.
How CoCo Works
A CoCo query travels through four layers before an answer comes back:
- 1
The user asks a question in natural language inside Snowflake.
- 2
CoCo parses the question and looks up the relevant metric, dimension, and filter definitions in Horizon Context, Snowflake's governed semantic foundation.
- 3
The query runs against Snowflake Semantic Views the native construct where Horizon Context stores governed metrics and relationships.
- 4
CoCo returns an answer with the data, the definition it used, and a record of which governance rules were applied.
Because Horizon Context sits between the question and the data, CoCo doesn't guess at definitions or invent its own. According to Snowflake's product announcement, CoCo and CoWork together hit 83% accuracy on complex enterprise queries when paired with Cortex Sense, compared with 47% without that enrichment layer.
What CoCo Solves
CoCo closes the gap between a business question and a trusted Snowflake answer, without the user needing to write SQL, find the right table, or remember how a metric was defined six months ago.
Natural-language access to governed data. Business users ask questions in their own words and get answers that match the data team's approved definitions.
Governance applied at query time. Row-level masking and role-based access flow through automatically, so a CoCo answer for a regional finance lead and a CoCo answer for a global VP can return different views of the same metric based on what each user is allowed to see.
Audit trails on every answer. CoCo records which definition produced the result, which is the kind of evidence finance, audit, and compliance teams need before relying on AI for decisions.
Where CoCo Stops
CoCo answers questions inside Snowflake. It doesn't natively answer questions inside Power BI, Excel, Tableau, Looker, ThoughtSpot, or any external AI agent like Claude or ChatGPT. Most enterprises run the business in those tools, not in Snowflake's native AI surface, and that's where the trust gap reappears even after CoCo is turned on.
A CoCo answer for "Q3 revenue" and a Power BI dashboard for the same metric can disagree, because the BI tool doesn't natively read Snowflake Semantic Views or the definitions Horizon Context governs. The same problem shows up when an external Claude or ChatGPT agent queries the data without going through a layer that enforces those definitions.
The AtScale Snowflake Semantic Views guide describes the gap directly: "Snowflake Semantic Views centralize governed metric definitions inside Snowflake, but Microsoft tools like Power BI and Excel do not natively consume those definitions."
How to Extend CoCo Answers to Every Tool the Business Uses
A universal semantic layer like AtScale for Snowflake reads the same Snowflake Semantic Views that Horizon Context governs and exposes those governed definitions to Power BI, Excel, Tableau, Looker, Google Sheets, and external AI agents through standard interfaces (XMLA, MDX, DAX, JDBC, MCP). The result is one definition of revenue, one definition of active customer, one definition of gross margin, computed the same way whether the question comes from CoCo, from a Power BI dashboard, from an Excel pivot, or from a Claude agent.
One definition of every metric. CoCo, Power BI, Excel, Tableau, and AI agents — all reading the same governed source.
What the Anthropic Benchmark Says About CoCo's Ceiling
The accuracy question stopped being theoretical when Anthropic published its own numbers. In How Anthropic's AI Accuracy Went from 21% to 95%, AtScale CTO Dave Mariani walks through what happened when Anthropic's data science team ran internal analytics on Claude. Given raw access to thousands of SQL files, dashboards, transforms, and notebooks, Claude answered correctly about 21% of the time. After Anthropic added a semantic layer and a rule telling Claude to consult the layer before guessing, accuracy moved to roughly 95%, with some domains reaching close to 99%.
AtScale's production benchmark with the commercial banking arm of a Tier 1 global bank pushed the result further. Routing queries through the AtScale semantic layer cut compute by up to 21,000x and lifted accuracy from around 70% to 100% on a set of five common questions, with projected savings of roughly $9 million a year on that one question set. CoCo's accuracy is bounded by the quality of the semantic layer feeding it — Horizon Context gets CoCo to a strong baseline inside Snowflake, and a universal semantic layer carries the same accuracy out to Power BI, Excel, Tableau, and external agents.
After adding a semantic layer to Claude's analytics workflow.
Accuracy from ~70% to 100%, $9M/yr projected savings.
Frequently Asked Questions
Is Snowflake CoCo the same as Cortex Analyst?
CoCo is Snowflake's branded conversational AI surface for business users, built on top of Snowflake's Cortex foundation and integrated with Horizon Context for governed business meaning. Cortex Analyst is the underlying capability, and CoCo is the productized experience.
Does CoCo work outside Snowflake?
No. CoCo answers questions inside the Snowflake environment. For governed answers in Power BI, Excel, Tableau, or external AI agents, the same definitions need to be served by a universal semantic layer that reaches those tools.
Does CoCo need Horizon Context to work?
CoCo can return answers without governed definitions, but the accuracy and trust profile drops sharply. Horizon Context is what makes CoCo's answers grounded in the data team's approved logic instead of in the model's guess.
Can CoCo answer questions for AI agents, not just human users?
Yes. CoCo can be consumed programmatically by AI agents through Snowflake's APIs, which makes it a governed query layer for agent workflows, not just an interface for business users.
What's the difference between CoCo and CoWork?
CoCo answers questions. CoWork takes action across multi-step tasks. They share Horizon Context as the semantic foundation.
Related Terms
Key Takeaways
Snowflake CoCo is the conversational AI assistant for Snowflake data, built on Horizon Context for governed business meaning.
CoCo's accuracy depends on the semantic layer feeding it, and the Anthropic benchmark shows the difference between a raw model (21%) and a semantic-layer-grounded one (95–99%).
Horizon Context closes the meaning gap inside Snowflake, and a universal semantic layer like AtScale closes it everywhere else the business asks the same question.