Claude Dashboards + Snowflake: A CFO Dashboard in 1 Prompt, No Power BI, No Code

DATA ENGINEERING

How do you build a CFO dashboard in Claude without Power BI or code?

In short: Claude Dashboards, connected to Snowflake via an OAuth connector and MCP server, turns one plain-English prompt into a finished CFO revenue dashboard in roughly 10 minutes, with no Power BI, no BI ticket and no code.

Ani Björkström, the Stockholm-based consultant behind the YouTube channel "Ani Björkström | AI for Finance," demonstrates Claude Dashboards — in beta since 8 October 2026 — connected to a live Snowflake data warehouse. Rather than filing a ticket with a BI team and waiting weeks for a Power BI report, a business user types one prompt describing the dashboard they want, and Claude builds it directly against the company's own data.

What do you need to connect Claude to a Snowflake data warehouse?

Connecting Claude to Snowflake requires a Snowflake-managed MCP server and the Claude Snowflake connector, authenticated through OAuth. A technical user sets this up once, and it is not a complicated process.

The setup involves creating the MCP server with the queries needed to run it, establishing an OAuth user, then in Claude going to Customize → Connectors → Add connector. Once that connector exists, Claude can read directly from the company's Snowflake tables — in this case a table called "monthly revenue" containing net revenue in USD and order counts broken out by month, region, country, nation and customer segment, spanning January 1992 to July 1998.

What does the one prompt for a CFO revenue dashboard actually say?

The prompt describes the data source and the business requirement in plain language: it names the Snowflake table, explains that it holds net revenue and orders by month, region, country, nation and customer segment, and asks Claude to "build an interactive Claude dashboard for CFO called Snowflake revenue dashboard," specifying exactly which figures should appear.

Ani Björkström notes that this is where a business user has an advantage over a technical one: the business user knows the report specification and which figures matter, without needing to translate that into a developer ticket. After hitting enter, the dashboard took about 10 to 11 minutes to generate.

What did the finished Snowflake revenue dashboard show?

The finished dashboard displayed net revenue, orders for the same period, average order value and the last month in the dataset, followed by a chart of net revenue by month showing every dip and peak across the time span. Below that, it broke down year-over-year change by region and by customer segment.

Between 1996 and 1997, Europe recorded the steepest decline at minus 20%, with the Middle East, Asia and Africa also down while America was the only region to grow. By customer segment, every category declined except automobile. A final table listed net revenue by country, region and order count, with a search function to filter to a single country.

Can you trust an AI-built dashboard, and what's the Snowflake trial trap?

Every number on the dashboard is clickable, revealing the exact SQL query and result used to calculate it, which can be copied into Snowflake directly to confirm the figure matches. Clicking refresh pulls updated numbers straight from the warehouse, since the dashboard queries live data rather than a static export.

One setup detail catches people out: Claude cannot connect to a Snowflake trial account. The connection must be made to a real company Snowflake account, either set up directly or by asking a technical team to configure it once. Finished dashboards can be shared via a private link, opened by inviting colleagues, exported to tools like Hex, or turned into board-ready slides with a follow-up request to Claude.