Claude Dashboards: Connect Data Sources, Inspect SQL and Verify Results
Direct answer
Claude Dashboards is a paid-plan beta for exploring connected business data. It builds charts with inspectable queries and refresh timestamps, but you still need to verify what each metric counts. Pro, Max, Team and Enterprise are eligible; Free does not include the Dashboards template. Enterprise owners must enable it first. Official getting-started guide.
AI Tool Finder Editorial Team · Sources checked October 9, 2026
This is a documentation-based guide, not a hands-on product test. Vendor capabilities are attributed below. Examples and verification steps are our editorial proposals; no account was connected and no product result is claimed.
Access and supported data sources
Documented warehouses include Amazon Redshift, BigQuery, ClickHouse, Databricks and Snowflake; connected apps can also supply data, with Salesforce named as an example. Ask in chat, select Output → Dashboards, or start in Artifacts. Each chart exposes SQL and its last-refresh time. Usage counts against your Claude plan. Sources, creation and usage.
Artifacts require cloud code execution and file creation. Check personal Capabilities or the organization’s settings as appropriate. On Enterprise, an Owner can restrict Dashboards to selected groups through custom roles. Artifact prerequisites · Organization controls.
Eligibility is not a completed connection. Confirm the exact account, tables, role and data permissions with your data owner before granting access. We have not connected a warehouse or CRM, measured query cost, or verified a refresh interval.
A workflow that makes a chart reviewable
- Agree the question. Name the decision, the reporting period, the population and the exclusions. “Revenue this month” is incomplete until currency, refunds and recognition rules are defined.
- Choose the minimum source. Prefer a curated, permissioned view containing only the fields needed. Ask the data owner whether the query can expose personal or confidential information.
- Create the exploratory dashboard. Describe the question, name the source and request explicit metric definitions beside each chart.
- Open every chart’s query. Read its selected fields, joins, filters, date boundaries and aggregation. Ask for a correction if the query answers a different question.
- Reconcile against a trusted result. Compare one small period and a known subset against the source system or an approved query. Record unexplained differences.
- Review freshness and audience. Record the displayed refresh timestamp and confirm who may see the result. Share only after the data owner accepts the scope.
These are our proposed review steps. An attractive dashboard is an exploration artifact; it does not replace your metric definitions, reconciliation or access review.
Inspect SQL, filters and freshness
Swipe wide tables horizontally on a phone.
| Check | Typical ambiguity | Acceptance question |
|---|---|---|
| Counting unit | Rows, users, accounts and orders are different. | Is the query counting the agreed unique key? |
| Join cardinality | One customer can have many orders and many events. | Does a join multiply rows before aggregation? |
| Period and time zone | Creation, payment and event time can differ. | Which timestamp and time zone determine the bucket? |
| Status and exclusions | Test accounts, cancelled orders or deleted leads. | Are inclusion and exclusion rules visible in the filter? |
| Currency and revenue | Gross sales, cash collected and recognized revenue. | Are tax, refunds and currency conversion treated consistently? |
| Freshness | Chart refresh time versus upstream ingestion lag. | When did the source last receive complete data? |
| Missingness | Null or unavailable data shown as zero. | Does the dashboard distinguish no data from a genuine zero? |
Check filter scope across charts: a region selector should not silently filter signups while leaving revenue global. After changing a filter, re-open the query or otherwise confirm the calculation changed as intended. Save the agreed query and definitions with the review evidence.
“Last refreshed” is not a guarantee of continuous streaming or complete upstream data. Ask for the actual refresh mechanism and source latency before promising a live operational report. No fixed refresh cadence or free warehouse-compute allowance is asserted here.
Three original dashboard briefs
These fictional requirements are not our Snowflake, BigQuery or Salesforce results. Replace source names only after access and metric definitions are approved.
1. SaaS signups, active users and revenue
From approved account, event and billing views, chart weekly signups, weekly active users and net collected revenue for 12 completed weeks. Define an active user by the approved meaningful-event list. Exclude internal/test accounts. Keep currencies separate. Show the query and freshness for each chart; list unresolved joins or definitions before presenting a combined funnel.
Review focus: a person may belong to multiple accounts, and a signup is not a paid customer. Reconcile a known week independently. Do not compare a user count with an account-level revenue denominator without explaining the mapping.
2. CRM opportunities by stage
Using the approved CRM opportunity source, show open opportunities by current stage and owner as of the displayed refresh. Include count and unweighted value separately. Exclude deleted and closed-lost records. Show unknown stage and missing owner as explicit categories. Do not describe this snapshot as historical stage movement.
Review focus: current stage is not a stage-transition history. Weighted pipeline requires a separately agreed probability model. Check one owner’s totals against CRM records with matching permissions.
3. Ecommerce orders and repeat purchases
Using approved order and customer views, show completed orders and repeat-customer share by month. Define repeat as at least one earlier completed order before the current period. Exclude test and cancelled orders; show refunds separately. State guest-customer identity limitations. Do not expose names, addresses or emails in the dashboard.
Review focus: counting two orders within one month is different from returning in a later month. Agree the identity rule and historical lookback before accepting a retention claim.
Private first; sharing is a separate permission
Dashboards begin private. In Team/Enterprise, outside sharing depends on owner policy. Connected-app artifacts cannot simply use an unrestricted “Anyone with the link” route; the share menu and artifact restrictions matter. Viewers use their own source connections, so missing source access can produce errors. External invitees cannot use connector-backed parts. Stored information can still be visible to people who open an artifact. Sharing and connected-data rules.
Our recommendation: review the exact artifact type and audience with an authorized teammate before sharing a link. Do not assume that because the chart opens for you it opens safely or completely for someone else. Minimize customer-level data, confirm retention requirements and remove access when the collaboration ends.
A paid Claude plan does not remove warehouse resource limits or provider charges. Ask the data owner about query scanning, concurrency and throttling. The reviewed documentation does not establish a universal query quota, storage allowance or refresh service level for every connector.
Docs, Slides and Design have different access
Anthropic’s October 8 announcement removes the beta label from Docs, Slides and Design and includes Free. Dashboards remains paid beta. For Enterprise, Docs/Slides/Design default enablement is October 15, 2026, with early owner enablement available. Launch announcement · Exact Enterprise defaults.
These are Artifact templates in Claude, not the Claude for Google Docs/Sheets/Slides add-on. Do not use template eligibility to infer Google Workspace sidebar access. Read the separate Workspace installation and approvals guide.
Choose this route for exploration
Good fit: an analyst and business owner exploring a bounded question with known definitions and permissioned data. Poor fit without more validation: a regulatory report, a customer-facing SLA dashboard or a replacement for a governed BI model.
If a number differs from your source report, stop the comparison and inspect definitions, joins and timing before changing the chart. If a connection fails, check source permissions and organization controls; do not upload an unrestricted export to bypass a denied connection.
Continue with Claude’s product profile, data-analysis tool choices or the productivity guide. Next step: agree one metric and one completed week, then reconcile that small result before extending the dashboard.