How to build a useful business dashboard with Claude—and check its numbers

A dashboard earns its place when it helps you make a decision. For a small service business, that might be: Which opportunities need attention this week? Where are deals getting stuck? How much potential work is sitting in the pipeline?

Claude can help build that first view. You still need to decide what the numbers mean and check that the dashboard counts the right things. Here is a manageable way to start.

1. Check the prerequisites

Anthropic introduced Claude Dashboards on October 8, 2026. It connects to supported sources such as Salesforce and data platforms including BigQuery and Snowflake. It is designed for exploratory analysis, with a visible query and refresh information behind each chart. Read the announcement.

As of October 19, the feature is in beta on Pro, Max, Team, and Enterprise. Enterprise owners must enable it. You also need access to a compatible source containing the relevant data. A paid subscription alone does not supply that data or its permissions. Current setup guide.

If your business runs entirely in a local spreadsheet, do not assume an uploaded file becomes an automatically refreshing dashboard. Start by defining the report and checking a dated export. Investigate supported connections before committing to a live workflow.

2. Choose one decision and three metrics

Use a hypothetical consulting business with opportunities in Salesforce. Its owner wants to decide whom to follow up with on Monday.

Start with three metrics:

• Open opportunity count: distinct opportunities currently in an active sales stage, excluding won and lost deals.
• Open pipeline value: the sum of the amounts on those same opportunities, in one agreed reporting currency. This is potential business, not booked revenue.
• Stale opportunity count: open opportunities whose latest recorded sales activity was more than 14 days ago. Show records with no activity date separately.

That last definition needs care. Does “activity” mean a customer email, a completed call, or any edit to the record? Ask the person maintaining the CRM. A changed phone number should not necessarily reset the follow-up clock.

Keep the first version small. You can add another chart after the first three prove useful.

Help me define a Monday follow-up dashboard for a small consulting business. Use three metrics: open opportunity count, open pipeline value, and stale opportunity count. Before building, ask me to confirm active stages, reporting currency, the activity field, the 14-day rule, and the timezone. Flag ambiguous definitions. Do not guess missing business rules.

3. Connect only the data you need

In Claude, open Customize > Connectors, find the relevant service, review its capabilities, and complete its connection flow. Team or Enterprise organizations may require an owner to enable the connector first. Review available tool permissions and prefer read-only access for this analysis. Connector instructions.

Have your CRM administrator confirm which fields you can read. Use a restricted reporting account or permissions where appropriate. A prompt saying “do not edit records” is useful direction; access controls enforce the boundary.

First ask Claude to identify available fields and explain their meaning. If it cannot access the activity history needed for your definition, stop that metric rather than quietly substituting another field.

4. Build a plain first version

You can ask for a dashboard in chat, choose Output > Dashboards, or start from a Dashboards template in Artifacts. Refine it through follow-up requests. Creation steps.

Build a dashboard from my connected Salesforce opportunities using the definitions we confirmed. Include three metric cards and a table of opportunities needing follow-up, with owner, stage, amount, and last activity date. Put missing activity dates in a separate group. Explain the source fields, filters, and calculations. Do not change CRM records or share the dashboard. If a required field is unavailable, tell me instead of inventing it.

Use ordinary labels. “Open pipeline value” is clearer than “Revenue opportunity engine.” Put the currency and reporting scope where the reader can see them.

5. Reconcile the numbers before trusting the view

Pick a small slice you can check yourself: one owner and one stage. Run a source-system report with the same filters and compare distinct opportunity IDs, counts, and amounts.

For example, suppose the source report lists ten opportunities totaling $42,000, but the dashboard shows fourteen and $59,000. That difference could come from counting one opportunity repeatedly after joining it to several activity records. It could also be a filter mismatch. Inspect before choosing a fix.

For owner [name] and stage [stage], reconcile the dashboard against this source report: [provide an authorized, minimal export]. Compare opportunity IDs, distinct counts, and summed amounts. List unmatched records. Check for one-to-many joins, excluded stages, currency differences, and missing values. Explain the discrepancy before changing the query. Do not force totals to match by dropping unexplained rows.

Open the chart's query and ask for a plain-English explanation. If you cannot assess it, have someone familiar with the data review it. A plausible explanation from the same assistant is not independent verification.

6. Test the awkward cases

Check an opportunity with several activities, one with none, a recently closed deal, and a record near the 14-day boundary. Confirm whether “more than 14 days” uses calendar dates or elapsed hours.

Then check freshness. Read the displayed refresh time and compare a known source change after refreshing through the supported interface. Do not label a dashboard “real time” based only on its appearance.

Before sharing, review who can open it and what records they will see. Name a person responsible for checking it when stages, fields, or business definitions change.

Make it part of a useful routine

Spend ten minutes on Monday reviewing the verified follow-up list. Let a person decide the next action; the dashboard does not need permission to email prospects or change deal stages.

If your process needs a custom intake or tracking app, see building your own business apps. For keeping human checkpoints in a recurring workflow, see the AI blog publishing system guide.

The win is a small report you understand well enough to use. If you want help choosing the first workflow or checking its setup, get in touch.