{"slug": "snowflake-coco-for-excel-and-google-sheets-governed-ai", "title": "Snowflake CoCo for Excel and Google Sheets: Governed AI", "summary": "Snowflake launched Snowflake CoCo, its coding agent for data teams, inside Microsoft Excel and Google Sheets, letting business users build workbooks with native formulas, charts and pivot tables through natural language while querying live Snowflake data under their own role-based permissions, masking and row access policies. CoCo runs queries with the user's permissions instead of requiring CSV exports, ODBC drivers or third-party connectors, and saves query text into workbook settings so data can be refreshed manually via a button or the /refresh skill, or automatically when a workbook is opened. Snowflake said the integration draws on Semantic Views, Business Ontology and Snowflake Cortex Sense to keep AI-driven analysis consistent and accurate.", "body_md": "We are excited to bring Snowflake CoCo, the coding agent for data teams, to two of the world’s most used analytical surfaces: Microsoft Excel and Google Sheets. These offerings natively integrate governed Snowflake data, semantics and AI within the spreadsheet interface analysts already use. Business users interact with CoCo in natural language to build workbooks with native formulas, charts and pivot tables — without worrying about mundane cell formatting and data integration. As agents become the interface for building spreadsheets, users can accomplish more with greater efficiency with the tools they already use.\n\n## Manual exports and connectors are painful\n\nSpreadsheets have been the dominant interface for interacting with data for over 40 years. Despite this ubiquity, teams find having consistent data across their spreadsheets challenging — with file names, comments and emails marking different versions and sources. Typically, users have three options to connect data:\n\n- **CSV exports** : Spreadsheet users download query results, tables or notebook outputs locally to their device.\n- **ODBC connectors** : Spreadsheet users ask their data team to configure ODBC drivers to connect to their data using SQL.\n- **Third-party connectors** : Spreadsheet users ask their IT team to provision and maintain a separate system for connecting data.\n\nManual exports result in disconnected copies of data, creating data silos. For administrative teams, this creates governance challenges across the data sprawl. For business users, this results in inconsistent analysis and isolated work.\n\n### AI is an expectation for productive work\n\nCrucially, the limitations of manual data connectors make it difficult to integrate AI into spreadsheets. We anticipate that agents will become the default mode for working with spreadsheets, similar to other interfaces such as the coding integrated development environment (IDE). AI can handle the rote parts of building a workbook, such as pulling and cleaning data, writing complex formulas and creating models — freeing the analyst to focus on analysis.\n\n## Meet Snowflake, with AI, now in your spreadsheet\n\nSnowflake CoCo for Microsoft Excel and Google Sheets bring Snowflake to the spreadsheet where analysts live, without requiring them to learn an entirely new and different tool. CoCo creates and runs queries against Snowflake data with the user’s permissions, removing the need for traditional connectors. CoCo then builds workbooks, using cells as context to create native spreadsheet objects.\n\n## Live, governed Snowflake data with intelligent refresh\n\nCoCo for Excel and Sheets deliver live Snowflake data directly to your spreadsheet while [maintaining enterprise governance](https://www.snowflake.com/en/blog/governed-ai-enterprise-controls-snowflake-coco/) and context. After users sign in, CoCo uses their selected or default Snowflake role to limit their access. [Snowflake access controls](https://www.snowflake.com/en/blog/engineering/rbac-simplified-inherited-grants-access-requests/), masking and row access policies apply to the results, which are written to your spreadsheet. These data and AI queries benefit from all of your Snowflake context, including Semantic Views, Business Ontology and Snowflake Cortex Sense which are designed to provide analysts with consistent, accurate results — crucial when using AI for trustworthy analysis.\n\nAfter data is pulled into a spreadsheet, intelligent refresh makes it easy to keep data up to date. When CoCo executes queries to pull data, it saves the query text into the workbook’s settings. As CoCo refreshes data, it can intelligently update the workbook to account for changing result sizes. Users can trigger refresh in two ways:\n\n- **Manually** : This is done via a button or the`/refresh` skill. This is intended for analysts to update their models with the latest data.\n- **Automatically** : This is done when the workbook is opened. This is intended for executives to open a workbook and always have the latest data, without needing to click or configure anything. This is supported for CoCo for Excel only.\n\nIf the query changes, such as for a different result size or a different set of columns returned from a SELECT * statement, CoCo updates the workbook to keep existing data, formulas and assets intact.\n\nSince all queries are saved into the workbook, users have clear visibility into what data assets were used for analysis. This makes it easy to reconcile different answers across workbooks, finally answering the question of which number we can trust.\n\n## CoCo in the workbook unlocks natural language analysis, agentic manipulation and AI functions\n\nCoCo brings agentic workflows to manipulate spreadsheets directly, going beyond chatbot integrations and data connectors. Instead of answering questions in a side panel, CoCo generates output natively in the spreadsheet with formulas and objects. In Excel, users can call supported Snowflake Cortex AI functions directly in spreadsheet cells, such as `CORTEX.CLASSIFY` and `CORTEX.SUMMARIZE`.\n\nFor example, you can ask CoCo to show quarterly revenue by region and growth projections on your sales data. In seconds, CoCo translates that into a plan to build an analysis, complete with formulas and pivot tables:\n\nThese agentic capabilities extend to the surface in which analysts live with the core analytical objects that spreadsheet users build. Since CoCo has access to your skills, your workbook and your data as context, it can build the way a human analyst would:\n\n- **Visualizations** : CoCo analyzes the shape of data to select the most effective chart type. Charts remain fully live, interactive and connected to the spreadsheet for easy editing.\n- **Pivot tables:** CoCo uses Snowflake compute for the heavy lifting of data modeling and aggregations to build pivots, even for source data sizes that exceed Excel’s 1 million-row limit and Sheets’ 20 million-cell limit.\n- **Formatting and validation:** CoCo automatically applies conditional formatting, precise number formats and data validation rules. These can be customized with skills specific to your workflow and practices.\n\nThe Chefs’ Warehouse found that CoCo for Excel’s agentic capabilities made handling their complex tabular data easy and intuitive:\n\nCoCo for Excel’s performance and capabilities are impressive. Our data includes many complex tables, and CoCo understands the relationships and structure to execute the right joins and build workbooks. Best of all, it’s intuitive, so our analysts can use it without training.\n\n##### Matt Wyman\n\nVice President of Information Technology, The Chefs’ Warehouse\n\n## Natively integrated with the Snowflake data platform, including context\n\nCoCo tightly integrates with your context in Snowflake, including Semantic Views, Business Ontology and Cortex Sense. Using Snowflake context, CoCo grounds analysis in your company’s Business Ontology to understand entity relationships and definitions, with Semantic Views to connect concepts with data. Additionally, CoCo integrates with Cortex Sense to automatically cover the long tail of data assets, which are particularly useful for ad hoc analysis. Many teams already curate context for their AI and BI applications, which now extends directly to Excel and Sheets.\n\nAdditionally, CoCo can offload intensive, complex calculations to Snowflake’s scalable virtual warehouses. Traditionally, for large data sources, users would ask their data team to build pipelines to curate custom data products. With CoCo for Excel and Sheets, complex queries run inside a Snowflake warehouse instead of directly within the spreadsheet interface in the client. This handles work that exceeds what the spreadsheet interface can hold or calculate.\n\n## CoCo is built for spreadsheet work with Snowflake data\n\nCoCo is [purpose-built for efficient data work](https://www.snowflake.com/en/blog/engineering/data-eng-bench-data-engineering-agent-benchmark/). We extended that foundation into Excel and Sheets with tools for inspecting workbooks, writing formulas, formatting cells and creating charts and pivot tables.\n\nHowever, modeling actions are only part of the workflow. Answering a business question also requires access to and understanding of underlying data. CoCo combines Snowflake object discovery and access to semantic information with workbook-specific tools, helping users move from a question to relevant data and analysis.\n\nMany AI tools can work in a spreadsheet, but CoCo for Excel is built for our work and starts with our data directly. There’s no exporting or switching of tools; it gets to the right answer faster, and our teams spend less time rebuilding reports.\n\n##### Kate Savina\n\nHead of Data Technology, GTCR LLC\n\nTo evaluate the end-to-end spreadsheet workflow, we internally tested common spreadsheet tasks in Excel that included reasoning, modeling and pulling Snowflake data.<sup>[\\[1\\]](#footnote-1)</sup> We measured performance across three dimensions (accuracy, latency and output tokens per task) for both CoCo for Excel and an AI model harness for Excel using the Snowflake MCP, across three trials. The add-ins had similar accuracy on these tasks, but CoCo for Excel had better latency and used fewer output tokens:\n\nCoCo for Excel is optimized for working with data, including object discovery, understanding and scale. CoCo orchestrates direct-to-sheet data transfers, which stream query results directly into the workbook without using the model’s context window. This architecture is designed for large-scale data manipulation and analysis, leveraging Snowflake’s compute capabilities and asset integrations:\n\nThese internal results reflect CoCo’s design goal as the coding agent for data teams — maintaining accuracy with improved latency and fewer output tokens per task. Data context grounds the analysis, while optimized data transfer reduces the model work needed to deliver it.\n\n## Use Snowflake CoCo for Excel and Sheets today\n\nSpreadsheet users can now discover greater insights and perform deeper analysis with Snowflake CoCo for Microsoft Excel and Google Sheets — an easy, trusted and efficient AI for spreadsheets. Both are now available in public preview. To get started:\n\n1. Install CoCo for [Excel](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code-excel-integration) or CoCo for[Sheets](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code-sheets-integration) using our documentation\n2. Log in to your Snowflake account to start your first analysis\n\nStay tuned for additional feature updates, such as adding spreadsheet lineage to Horizon Catalog, exporting visualizations and more.\n\n*This article contains forward-looking statements, including about our future product offerings, and are not commitments to deliver any product offerings. Actual results and offerings may differ and are subject to known and unknown risk and uncertainties. See our latest 10-Q for more information.*\n\n1.  Internal tests across models and harnesses were run between Oct. 26, 2026, and Oct. 28, 2026, using Excel for Web. Results reflect 25 spreadsheet tasks run with Opus 5.5, Sonnet 5.5 and Haiku 5.5 in both add-ins, across three trials; Figure 5 shows median seconds per task across the trials. In Figure 6, tasks that failed or timed out are shown at the 900-second cap. [↩](#footnote-ref-1)", "url": "https://wpnews.pro/news/snowflake-coco-for-excel-and-google-sheets-governed-ai", "canonical_source": "https://www.snowflake.com/content/snowflake-site/global/en/blog/snowflake-coco-microsoft-excel-google-sheets", "published_at": "2026-11-05 16:54:07.442030+00:00", "updated_at": "2026-11-05 16:54:09.748486+00:00", "lang": "en", "topics": ["ai-agents", "ai-products", "artificial-intelligence", "ai-tools"], "entities": ["Snowflake", "Snowflake CoCo", "Microsoft Excel", "Google Sheets", "Snowflake Cortex Sense", "Semantic Views", "Business Ontology"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/snowflake-coco-for-excel-and-google-sheets-governed-ai", "markdown": "https://wpnews.pro/news/snowflake-coco-for-excel-and-google-sheets-governed-ai.md", "text": "https://wpnews.pro/news/snowflake-coco-for-excel-and-google-sheets-governed-ai.txt", "jsonld": "https://wpnews.pro/news/snowflake-coco-for-excel-and-google-sheets-governed-ai.jsonld"}}