On this page
Key takeaways
- Dashboards answer known questions on a schedule. Exploration starts with a new question each time.
- A grid makes follow-ups awkward: every new question means a new chart and a new place to put it.
- On a canvas, answers are objects linked to the question they came from, so the reasoning stays visible.
- Keep Superset dashboards for monitoring and use a canvas for investigation.
What dashboards are good at
Apache Superset dashboards do one job very well: show a fixed set of metrics, refreshed on a schedule, to many people at once. Weekly revenue, active users, pipeline by stage. Everyone looks at the same tiles and knows where to find them.
That is monitoring, and a grid is the right layout for it.
Where dashboards struggle
Monitoring raises questions. Revenue dropped this week. Why?
To answer that on a dashboard, you:
- Open the chart builder and create a new chart broken down by region.
- Find space on the dashboard, or open a new one.
- Build another chart by channel for the weakest region.
- Build another for the weeks before and after.
Each step is a new chart, a new saved object and a new place to look. The line of reasoning, "revenue fell, it was the West, it was paid search, it started in week two", exists only in your head.
A grid is built for answers you already know you will need. It has no natural place for the next question.
What a canvas does differently
On the DashboardGPT canvas, each answer is an object: a chart, a metric, a table or a written insight. The canvas is infinite, so there is always room for the next answer beside the one it came from.
| Dashboard grid | Infinite canvas | |
|---|---|---|
| Best for | Monitoring known metrics | Investigating new questions |
| How a chart is made | Build and save it | Ask a question |
| Follow-ups | New chart, new tile | Select an answer and ask |
| Layout | Fixed grid | Free: group, arrange, branch |
| Reasoning | Lives in your head | Visible as linked answers |
| Refresh | Scheduled | Scheduled on Team, or ask again |
Three ideas make the canvas work:
- Answers are objects, not tiles. Move them, group them, delete dead ends.
- The thread stays visible. Each follow-up is linked to the answer it came from.
- Group what belongs together. Headline, breakdown and cause sit side by side.
The same investigation on a canvas
- Ask "Weekly revenue for the last 12 weeks." A line chart appears.
- Select it and ask "Break the last week down by region." A bar chart appears next to it, linked.
- Select the West bar and ask "By channel, last 4 weeks." Paid search stands out.
- Ask "When did paid search orders start falling?" Week two.
Four answers, arranged left to right, with the reasoning visible. Share the canvas and a colleague can follow the investigation without a meeting.
Using both on a Superset stack
You do not have to choose. DashboardGPT is built on Apache Superset for query execution and charts, so the two fit together:
- Superset dashboards for the metrics everyone monitors every week.
- The DashboardGPT canvas for the questions those metrics raise.
A simple routine: review the dashboard on Monday; when something moves, open a canvas and investigate; share the canvas in the team channel with the conclusion at the top.
When to promote a canvas answer to a dashboard
If you find yourself asking the same question every week, it has become a monitoring question. Build it as a dashboard chart so it refreshes for everyone. The canvas is where questions start; the dashboard is where the permanent ones end up.