Kevin Stratvert turns a 500-order sales sheet into an interactive dashboard with one Copilot prompt, checks the totals by hand, then restyles it with a second prompt.
Build a Claude live dashboard you'll actually use, in 5 steps
Ben AI's 5-step framework for Claude's new live dashboards: set the one decision it serves, pick 5-10 metrics, check the data, design from references, then build, share and auto-refresh AI insights with a routine.
Ben AI · YouTube · published Nov 1, 2026 · Work & Operations
▶ Watch on YouTubePrompt
I want a Claude dashboard that helps me decide [one decision]. Interview me one question at a time until you can write its goal in one sentence. Then take my metric brain dump, suggest any I'm missing, and check you can pull each one from my connectors.
Visual breakdown
Claude's new dashboards are live, show the query behind every number and keep their own database. Ben's 5-step framework makes sure you build one you'll actually open: start from one decision, keep metrics few, verify the data, design from references, then build and schedule a routine for AI insights.
The problem: Most dashboards people build with AI never get used, usually because they're cluttered and nobody trusts or needs the numbers.
The 5-step dashboard framework
- ConnectorsTools like YouTube and marketing sources, plus warehouses (Snowflake, BigQuery) and CSV uploads
- 1. Set the goalClaude interviews you until the dashboard's goal fits in one sentenceDropped: Extra decisions beyond one to three
- 2. Pick metricsBrain-dump yours, let Claude suggest more, keep the 5 to 10 that matterDropped: Metrics that don't help the decision
- 3. Check the dataClaude tries to pull each metric, defines it, and suggests chart type and time periodDropped: Metrics it can't get yet
- 4. DesignPaste 2 or 3 reference screenshots, compare three layout variations, pick one
- 5. BuildClaude builds the live dashboard artifact; you approve connectors on open
- Live dashboardShared by link; a scheduled routine refreshes AI insights each morning
InputsManual stepClaude does itResult
Design rules he gives Claude
- Biggest metric at the top left
- Minimal scrolling, see everything at a glance
- No tiny text, make it as visual as possible
- A short line under each metric explaining it
By the numbers
5steps in the framework
1 to 3decisions per dashboard
5 to 10metrics
100 MBdatabase, plus 1 GB of files
Build yours
Do these in one Claude chat, using the free prompts linked in the video description.
- Ask Claude to interview you, one question at a time, until it can write the dashboard's goal in one sentence
- Brain-dump the metrics you want, ask for suggestions, and keep the 5 to 10 that help the decision
- Run the data-check prompt: connect missing tools and correct how each metric is calculated
- Paste 2 or 3 dashboard screenshots from Dribbble, Pinterest or Mobbin and ask for three layout variations
- Pick one, then ask Claude to build it as a dashboard artifact and approve the connectors
- Share it by team artifacts or link, and ask for a routine (e.g. every morning at 7 a.m.) that writes fresh insights
Shared dashboards use the opener's connectors; to give everyone the same numbers, ask Claude to save the data inside the dashboard instead of loading it live.
Lessons from the video
- Too many metrics is the number one reason people stop using a dashboard.
- It's now worth building a dashboard for a question you only care about for a week.
- Define each metric: 'new members' can mean all sign-ups or only paying ones.
- Insights that need AI interpretation can't load live; they need a scheduled routine.
- Views and conversions can be unrelated: some of his biggest videos brought almost no members.
Jump to the part you need
Distilled from the video’s captions
How to try it
- Open Claude.
- Copy the prompt and fill in the [placeholders].
- Watch Ben AI's video for the full walkthrough and setup.
Related use cases
Skill Leap AI gives Claude Code a prompt and the path to about 400 memory and knowledge files, and gets a 3D castle with rooms, a memory wall and a bookshelf of clickable books.
Metics Media rents a small Hostinger VPS, runs Hermes Agent in the browser, connects a cheap DeepSeek Flash model and Telegram, then has the agent schedule its own Barcelona to London flight-price watcher.
Futurepedia picks one of 86 explainer styles, adds a topic, and pastes the copied prompt into Claude. Claude shows a storyboard first, then generates the whole explainer in code with sound effects.