Claude Dashboards: The One Chart a Business Owner Checks Every Day
Last Wednesday we talked about using Claude to produce weekly reports automatically. This week we go a step further: instead of waiting for a report every week, build a chart that shows the latest numbers whenever you open it. On October 8, Anthropic launched Claude Dashboards (beta), which makes a great practice project for small businesses. This post first summarizes the key points from the official materials, then shares the approach I recommend.
Today's Highlights: Claude Dashboards are live, and owners can ask about their numbers directly
According to Anthropic's announcement and Help Center, using Claude Dashboards is straightforward: connect your data source to Claude, then ask in plain language, such as "revenue by region this quarter." Claude writes the query, draws the chart, and refreshes it as the data changes.
Which data can it connect to?
Supported data warehouses currently include Snowflake, BigQuery, Databricks, Amazon Redshift and ClickHouse, and it can also read apps like Salesforce through existing connectors. Anthropic's examples include revenue by region this quarter, weekly sign-ups by plan over the past six months, and Salesforce opportunities by stage and owner.
Every number can be traced to its source
The design I find most noteworthy is that every number can be clicked to show the query behind it, and every chart shows when it was last updated. People who can't read SQL can simply ask Claude to explain how a number was calculated. The New Stack adds that an Anthropic spokesperson said dashboards follow users' existing data permissions and are visible only to their creator at first; once shared, queries run by default on each viewer's own connection.
Know the limits too
The feature is currently in beta and limited to paid plans such as Pro, Max, Team and Enterprise, with usage counting toward your plan's allowance. It's off by default on Enterprise and must be enabled by an admin. Anthropic also says plainly that it suits quick, exploratory questions, and recommends professional BI tools for in-depth analysis. The New Stack also cautions that being able to see a query doesn't mean the query is correct.
What if your company has no data warehouse?
Frankly, most small businesses in Taiwan don't have Snowflake or use Salesforce. A closer path is the Google Docs, Sheets and Slides integration VentureBeat reported on October 6 (a public beta for paid plans): Claude can write formulas, build pivot tables and native charts directly in Google Sheets, and by default every change needs your approval before it's applied.
VentureBeat also notes that for tasks like merging tables or cleaning messy data, Claude pulls the data range into Python to process it, then writes the results back to the sheet; and it can only see files your Google account can already open. For small companies with data scattered everywhere, this path has a much lower barrier.
Jesse's Take: Define three numbers clearly before building a dashboard
I've seen many companies build dashboards, and the reason they get stuck is rarely technical. It's usually that the definitions of the numbers were never made clear. So for this mini project, I recommend the following three steps.
Suggestion 1: Define three numbers before you ask Claude anything
A common scenario: when the owner asks how much revenue came in this month, accounting means invoiced, sales means signed contracts, and the store means cash received. None of them is wrong, but a dashboard can only use one definition. Write down three numbers you must check every week, such as revenue, gross margin and accounts receivable. For each, write one sentence on how it's calculated and where the data comes from, and put that at the very top of your instructions to Claude.
Suggestion 2: Pull your data into one spreadsheet first
Many companies have data scattered across LINE groups, each department's Excel files and the owner's head. There's no need to rush out and buy a data warehouse. Start by consolidating sales and payments into one Google Sheet, one transaction per row, with fixed column names that never change. At minimum you need five columns: date, customer, item, amount and sales rep. Once the data is clean, whether you later chart it with Claude in Sheets or switch to Claude Dashboards, you'll save most of the effort.
Suggestion 3: Reconcile every number at least once
The first time, every number on the dashboard must be checked by hand. Pick last month's already-closed figures, have Claude calculate them, and compare them with accounting's report. If they don't match, open the query or ask Claude to explain the difference. Only after you've reconciled it once will you dare use it to make decisions in a meeting. If a number keeps failing to match, it usually means the way data is entered at the source needs to be standardized first.
An honest caveat: the examples Anthropic shows all assume the data is already well organized. They're best-case demos selected by the vendor, with some selection bias. Your data will be messier at first. That's normal, so don't let it convince you that you can't do this.
A first step you can take today
Open the sales Excel file you use most, spend ten minutes writing down the definitions of three numbers, upload it to Claude, and ask it to calculate last month's results based on those three definitions and draw a chart. Reconcile it once with accounting. If it matches, you have the foundation of your first owner's dashboard. Run the same instructions again next month and you can start watching trends.
Further Reading
Related posts on JCXO
Original sources
Comments