Claude Dashboards on your data in one to two weeks

• Claude Dashboards switched on and governed: plans, permissions and connectors set up properly.
• One warehouse connected. BigQuery, Snowflake, Databricks and Redshift connect natively. If your data lives in PostgreSQL, SQL Server or Fabric, we bring it to one of those, or connect it through our FactSmith gateway, which is in development.
• Governed views: Claude gets a read-only login that sees only approved views of your data, never the raw tables, with personal details masked or left out.
• Data handling set up: the right Claude plan, a clear list of who your data passes through, and a POPIA clause in the engagement.
• Three to five starter dashboards on the questions your business actually asks.
• Your golden questions agreed and proven: the acceptance test for the whole engagement.
• Your team trained: how to ask, how to read the query behind a chart, and when to ask for help.

Fixed price, quoted after a short call. Your Claude seats and your warehouse stay in your name.

A dashboard is only useful once you know it is right. So before handover, your owner and your accountant agree about twenty questions and their correct answers, worked out from the books. Then we ask Claude, in plain words, with no hints.

Here is the worked example on Wide World Importers, Microsoft’s public sample company, loaded into BigQuery as raw sales tables: invoices, invoice lines, orders, payments and customers, exactly as an operational system keeps them.

Claude Dashboards answered all twenty correctly. On the three questions with a second fair reading, such as whether follow-up backorders count as new orders, it gave both answers and said which one it used.

Sample data, re-dated so it runs to today. Figures exclude tax. The same twenty questions are re-run each month on the care plan.

A demo shows that a chart appears. A golden-question scorecard shows that the number on it is the one your accountant would sign. It also catches the moment a data load breaks, a definition drifts, or a new system arrives, because the scorecard stops matching.

Your tables stay in your warehouse, but whatever Claude reads to build a dashboard passes through Anthropic: column names, sample rows and every query result. So we decide what it reads before it reads anything.

• A read-only login for Claude that sees only a set of approved views, each carrying your agreed definitions.
• Names, phone numbers and ID numbers masked or left out, unless you approve their use in writing.
• Your team’s own warehouse permissions pointed at the same views, so every viewer sees only what they should.
• A Claude plan whose terms rule out training on your business content, or training switched off where that isn’t possible.
• The data flow written down at handover, so your POPIA paperwork matches what actually happens.

Claude cannot leak a column it was never allowed to see. Background: Your data and Claude.

Most teams then want every dashboard to use the same definitions, and a monthly check. That is Trusted numbers for Claude.