Source: raw/How_to_Actually_Build_Dashboards_with_Claude_Like_a_Pro.md; 2026-11-09 addition: raw/Claude_s_New_Artifacts_Are_Amazing_Master_it_in_15_minutes.md (see November update)
Creator: Ben AI · URL: https://www.youtube.com/watch?v=vS-gi6o4FvY · Duration: 9:55 · Platform: YouTube · Uploaded: 2026-09-21
Ben AI shows how he built one daily business dashboard in Claude. It merges Stripe, YouTube, his newsletter, his community platform and Google Analytics into one page, and a scheduled skill refreshes it every morning. His point is that most Claude dashboards go unused because they show too much. So most of the process is about deciding what to leave out, checking the data can actually be pulled, and wireframing before anything is built. The video predates the first-party Claude Dashboards artifact type (2026-10-08) and uses a live artifact. The selection steps carry over to the new type unchanged.
Key Takeaways
- The filter for every metric: “would you actually do something differently because of this metric or number?” If not, leave it off.
- Cap it. Start with no more than 15 metrics; he settled on 14. “The biggest reasons dashboards don’t get used is because of overwhelm and too much information.” Add more later. In November he lowered the cap to five to ten; see November update.
- Merged numbers are the payoff. YouTube alone gives views. Joined with his other tools, the dashboard shows sales calls booked per video, which no single platform reports.
- Test the data before designing. Have Claude confirm it can pull every chosen metric through your current connectors. Two of his failed: Stripe wasn’t connected yet, and LinkedIn followers were hard to track, so he dropped that one.
- Wireframe in monospace first. Ask for the layout as a plain-text monospace sketch and iterate on that before any design is built. Text sketches are faster to change.
- Refresh with a skill on a schedule. Once the dashboard looks right, have Claude write a refresh skill “with everything that you learned today baked in” and schedule it daily. The link stays the same.
- How he uses it: bookmarked, checked “every morning for 5 minutes”.
The six steps
- Define the goal, and what doesn’t belong. Decide whether it is a whole-business, sales, finance or marketing dashboard. His opening prompt: “I want to build an analytics dashboard for my business. The goal is simple. Every morning I want to open one link and see everything important happening across the business in one page. Now before we build anything I want to work out what actually belongs on it.” He then listed his focus areas: general business metrics, retention and churn (his current priority), YouTube and newsletter performance, website traffic and sources, and community engagement.
- Select the metrics. Claude returns a long list of candidates. Be strict and pick at most 15 that link directly to decisions.
- Connect the data. Ben uses Composio as a “meta connector” that reaches most of his stack through one connector. He says it makes hard-to-connect tools such as Google Analytics easy. Set-up details are in Claude Connectors.
- Test every data point. Prompt: “Now using the connectors you have access to, can you please make sure that you can get all of these data points?” Fix or drop anything that fails.
- Wireframe, then build. Prompt: “Based on the data points we defined, lay out how a dashboard would look like in monospace so I can get a quick idea of how it would look like.” Apply his layout rules (below), add your brand guide, then build as either:
- a live artifact in Claude (his choice, because he is the only viewer), or
- a page on a live URL through the Vercel connector, which he describes as free, so anyone can open it. Expect a few rounds on the design and on wrong numbers.
- Keep it fresh. Prompt: “Dashboard looks good. Now can you create a skill that updates the dashboard every morning in the most efficient way possible with everything that you learned today baked in? Then schedule it for 8:00 a.m. daily.” Claude created a dashboard-refresh skill and a scheduled task that runs it.
November update: a tighter cap (added 2026-11-09)
In “Claude’s New Artifacts Are Amazing” (youtube.com/watch?v=Lodg0K3_zqk, uploaded 2026-11-09), Ben gives a five-step version of the same process for the new Dashboard artifact type. Two numbers changed:
- Decisions: the goal is now framed as “What decision should this dashboard help me make?”, kept to one to three decisions.
- Metrics: “five to 10 metrics at most”, down from the 15 he allowed in the September video above.
The other steps carry over: ask Claude to suggest metrics you missed, have it confirm it can get each one and how it will count it, and plan the design from two or three reference screenshots before building. The full comparison with Robo Nuggets’ method is in Claude Dashboards and Motion.
Layout rules he gives Claude
- Biggest, most important metrics at the top or top left.
- As little vertical scroll as possible, so most of it fits in one glance.
- One view; no tabs.
- Short descriptive labels. He notes AI tends to add long sentences describing the data.
- Compact overall, with logos or icons where they help.
Try It
- Write the one-sentence goal (“every morning, one link, everything important”) and your 3 to 5 current focus areas. Ask Claude for candidate metrics, then apply the “would I act differently?” test until 15 or fewer remain.
- Run the data-access test before you design anything. If an analytics tool has no native connector, try a middleware connector such as Composio (see Connectors).
- Ask for a monospace wireframe and fix the layout in text using the rules above.
- Build it as a live artifact, or as a Claude Dashboard if you’re on a paid plan after 2026-10-08. That type adds a per-number query view and last-refreshed times.
- Have Claude save what it learned as a refresh skill and schedule it before your working day starts.
Implementation
- Tool/Service: Claude (Cowork) live artifacts; Composio connector; optional Vercel connector for a public URL; skills plus scheduled tasks for the daily refresh.
- Setup: connectors for each data source (his: Stripe, YouTube, email newsletter, community platform, Google Analytics, mostly through Composio).
- Cost: not given beyond Ben calling the Vercel connector free.
- Integration notes: live artifacts already fetch fresh data when opened (Cowork Live Artifacts). The scheduled skill means the numbers are already up to date before he opens it. ^[inferred] He sells a dashboard-builder skill that runs these six steps in his paid community. It wasn’t reviewed.
Open Questions
- The numbers shown are fake. Ben says he changed them for the video, so the screenshots don’t show real performance.
- The full 14-metric list isn’t given, only the categories.
- Refresh skill contents. What “the most efficient way possible” means in his refresh skill isn’t shown: whether it rebuilds the artifact or only updates data.
- Live artifact vs Claude Dashboard. Whether the new Dashboard type can be refreshed by a scheduled skill the same way is untested here.
Related
- Claude Dashboards and Motion — the first-party Dashboard artifact type launched after this video
- Cowork Live Artifacts — Seven Dashboard Recipes (Eliot Prince) — ready-made dashboard prompts by business function
- Claude Connectors — Composio and other middleware routes for analytics tools
- Scheduled Tasks — the daily refresh mechanism
- Claude Cowork for Marketing — the “marketing control center” pattern with live artifacts plus scheduled tasks
- How Anthropic’s Marketing Ops Team Uses Cowork — the scattered-metrics problem from inside Anthropic
- Claude Artifacts — the underlying artifact system