Building Dashboards

Dashboard building with the Agent

The Agent builds Stackless dashboards from governed Semantic Models, from chart cards you have already validated, or from a plain-language request. Everything it builds starts as a private draft in the conversation's artifact workspace, so you can review it before anyone else sees it.

The best Agent-built dashboards are model-backed. If the right Transformation Model or Semantic Model does not exist yet, the Agent should say so and propose the model work first, rather than building on raw tables.

Start with the business purpose

Begin with the decision or workflow the dashboard should support:

Build a weekly executive dashboard for revenue, orders, AOV, active customers, and channel performance. It should help leadership see whether growth is healthy and which channel needs attention.

Then specify the audience, time period, metrics, breakdowns, filters, and any tabs or sections.

Ask the Agent to inspect existing models first

Which Semantic Models and Transformation Models can support this dashboard?

This avoids duplicate modeling work and keeps the dashboard tied to governed metrics.

A good dashboard request

Build a dashboard draft called "Marketing Performance" using existing governed models.

Include:
- KPI cards for spend, net revenue, ROAS, orders, and new customers, compared with the prior period.
- A weekly trend of spend and net revenue.
- A channel table with spend, revenue, ROAS, orders, AOV, and CPA.
- A date range filter (default last 90 days) and a channel filter that apply to every widget.

Before publishing, show me the model dependencies.

Use chart cards as stepping stones

If you are unsure about a visualization, ask for a chart card first:

Create a chart card of weekly net revenue by channel for the last 90 days.

When it looks right, click Copy to a new dashboard on the chart card, or ask:

Add this chart card to the Marketing Performance draft and add a KPI card for total net revenue.

Only chart cards can be copied into dashboards. To show table rows on a dashboard, ask the Agent for a table widget built on a Semantic Model.

What to ask the Agent for

Some dashboard features are set through the Agent (or the Advanced (JSON) editor) rather than the widget editor:

• Tabs: "Split this into Overview, Customers, and Marketing tabs."
• Filter scopes and defaults: "Make the date filter default to the last 90 days and apply it to every widget except the trend."
• Comparisons: "Compare each KPI with the prior period."
• Thresholds and sparklines: "Color AOV red below $150 and add a sparkline to each KPI."
• Number formats: "Show revenue as currency with no decimals." Check the result; if a format doesn't render as expected, set Decimals in Edit widget.

Refine the draft

You can keep refining the draft yourself in the artifact workspace:

• Edit layout to drag widgets into order and resize them, then Done.
• A widget's ⋯ > Edit widget to change its measures, dimensions, chart type, formatting, or comparison.
• Filters at the top of the draft work just as they will once it is published.

Selecting the draft in the workspace makes it the Agent's active context, so follow-ups like "add a tab for retention" apply to it.

Publish

When the dashboard is ready:

1. Ask the Agent to prepare the publish, or click the rocket icon (Publish to production).
2. Review the publish dialog. It lists any private model drafts the dashboard depends on; those are published with it.
3. Confirm. If the change affects live dashboards, you confirm a second time.

The Agent does not choose an audience. A dashboard published by a non-administrator is visible to that person's role; one published by an administrator is visible only to administrators until roles are assigned under Dashboard Access. See Dashboards.

Edit a published dashboard

Published dashboards are edited through a replacement fork:

Fork the published Executive Overview dashboard, add a Customer Retention tab, and show me the model dependencies before publishing.

Selecting a published dashboard you can edit in the artifact workspace also creates the fork. Publishing the replacement keeps the dashboard's role access.

Schedule exports

If your role has core/dashboards:schedule_exports, the Agent can create email schedules:

Email this dashboard as a PDF every Monday at 8:00 AM to maya.chen@northwind.example and jordan.patel@northwind.example.

Recipients must already have access to the dashboard. See Scheduled exports.

Common dashboard pitfalls

Building on raw tables

Raw tables are rarely stable business surfaces. Ask the Agent to use governed Semantic Models, or to create the missing Transformation Model first.

Missing grain

If you ask for customer-level revenue but the widget uses order-level rows, totals can duplicate. Ask the Agent to state the grain of each widget.

Unclear metric definitions

"Revenue" may mean gross, net, booked, or collected revenue. Ask the Agent to use the existing metric definition, and check the widget's Metric definitions drawer.

Partial periods

Trends that end in an incomplete week or month look like sudden drops. Ask for complete periods only, or a fixed range that ends at the last complete period.

Publishing too early

Drafts are for iteration. Publish only after dependencies, filters, metric definitions, and the intended audience are reviewed.

Useful dashboard prompts

• "Build a dashboard from these chart cards."
• "Add a date range filter that applies to every widget."
• "Explain the metric definitions used on this dashboard."
• "Trace all upstream models for this dashboard."
• "Which widgets may double-count because of joins?"
• "Schedule a weekly Excel export of this dashboard to the finance team."