Claude Dashboards for your business: data and integration scope
To introduce Claude Dashboards, separate three jobs: connecting an available source, preparing data that is not ready yet, and running a report other people depend on. If your team already has agreed metrics and authorized access, evaluate the native feature first. If departments calculate different numbers or someone still loads data manually, the scope includes integration and data validation. Identify that gap before deciding what to buy.
What the feature offers and what to verify
Anthropic announced Claude Dashboards on October 8, 2026. It positions the feature for exploratory questions alongside BI and analytics tools. An announcement establishes a stated capability; it does not demonstrate accuracy or savings in your operation.
The official Dashboards guide, checked October 20, describes a beta on Pro, Max, Team and Enterprise; Enterprise requires an owner to enable it. It documents inspecting each chart's query and last update. Example sources include BigQuery, Databricks and Snowflake, plus connected apps such as Salesforce. Verify your actual source and account before committing to a project scope.
A chart does not settle the metric definition. Monthly revenue could mean invoiced, collected or recognized revenue. Agree which number the reporting solution must produce and who may use it.
Identify the missing work before requesting a quote
This matrix is a scoping method proposed by Duotach, not a benchmark or a completed implementation.
| Current situation | First work to evaluate | Deliverable to request |
|---|---|---|
| Connectable source, clean data and an agreed metric | Native configuration and use | Reviewed query and independently checked result |
| Separate ERP and CRM without a shared identifier | Data integration and preparation | Join rule, exceptions and traceability |
| Two departments calculate the same metric differently | Business-owner agreement on its definition | Approved formula, exclusions and reference |
| Periodic report with recipients and a delivery time | Reporting operation and visualization | Agreed update, failure notification and owner |
If the missing piece is joining sources, another dashboard leaves that work unresolved. If the data is ready, a custom application may add operational work without addressing another requirement. Ask each proposal to separate reuse from development.
Define a metric with a reconcilable example
Write a specification for each metric: business question, source, unit of analysis, formula, period, time zone, currency and exclusions. Name the person who approves it and the record used to check the result. This becomes the acceptance contract for the number.
Fictional example: Operations wants dispatched orders while Finance wants collected revenue. An order with two payments can be duplicated when tables are joined. A return can change an amount after a period closes. Include those cases in the test and explain what each row counts before accepting the total.
Keep a reference calculated by the business owner separate from the dashboard result. Investigate differences in dates, status, duplicates and rounding. If both use the same unreviewed query, agreement provides little independent evidence.
Limit the connection and permissions
Inventory sources, necessary fields, authorized access, required update frequency and recipients. Begin with the smallest dataset that answers the question. A monitoring dashboard does not automatically need permission to change orders or collect payments.
Have IT test the complete path: who queries the source, what information remains in the artifact and what its recipient can see. Check for stored extracts and files too. Do not assume revoking source access removes copies that have already been shared.
Include an authorized test user and another without access to restricted data. Acceptance criteria should describe what each sees and what happens with insufficient permissions. Keep synthetic data until that separation can be demonstrated.
Test the number and the update separately
Agree on the test before development. Use a controlled copy with synthetic or authorized data; preserve the query version, data cutoff and expected result.
Initial number.
Reconcile each metric against the approved reference and explain differences.
Known change.
Add a test transaction and a correction. Verify what changes after an update.
Unavailable source.
Interrupt access in the test environment. The user must distinguish stale data from a real zero.
Permission change.
Repeat the query and artifact inspection with the agreed user profiles.
Absent owner.
Another team member should reproduce the check using the delivered documentation.
These are proposed project tests, not guaranteed Claude Dashboards behaviors. Test unattended delivery separately if it is required. A chart timestamp does not establish automatic delivery or error recovery.
What the supplier's proposal should separate
Compare scope using the same metric and sources. Request separate lines for data preparation, connection, visualization, tests, documentation and ongoing operation. Identify product licenses and data-platform consumption, with conditions checked for your account. There is no universal project price.
Define what happens when a column changes, an authorization expires or a business formula is updated. Agree who detects the issue, who fixes it and what support covers. Accepting a delivery and maintaining it are different commitments.
If you already use BI, keep the reference report during evaluation. Require evidence that the new path meets your requirements before retiring the system supporting your close. An exploratory test need not become a migration.
When to ask about integration
Duotach offers system automation and integration. Scope can include connecting sources and preparing an agreed reporting workflow. Claude Dashboards availability, system access and operational requirements are verified before defining the work.
Tell us which decision you need to make, where the data lives and which reporting steps are manual. Do not send credentials or confidential databases. An inquiry starts a scope conversation; discovery or development using company materials is agreed as paid work.
Define the connection and the metric
Describe the source, metric and manual step to discuss a bounded scope.