You already know which numbers matter in your business. The problem is that getting them means digging — exporting from the CRM, re-sorting a spreadsheet, asking someone who's busy. So you check the numbers you care about once a quarter, in a panic, right before a decision.
On October 8, Anthropic shipped a beta feature that changes the digging part: Claude Dashboards. You connect a data source, ask a question in plain English, and Claude writes the queries, picks the charts, and builds a dashboard that refreshes as your data changes. It's in beta on paid plans, and the supported connections are the enterprise-grade ones — BigQuery, Snowflake, Databricks, Redshift, ClickHouse, and Salesforce.
Most of the coverage is aimed at data teams and enterprise BI users. Here's the version for a business with no analyst on staff.
The feature that matters most is the smallest one
Every chart in a Claude dashboard shows the query behind it, along with the last refresh time.
That sounds like a technical detail. It's actually the trust mechanism, and it's the reason this is usable by someone who can't read SQL.
Dashboards usually fail for a boring reason: the owner can't tell whether the number on the screen means what they think it means. "Deals this quarter" — does that mean closed-won? Everything in the pipeline? Does it include the two renewals that got logged funny? When you can't answer that, you don't trust the chart, and when you don't trust the chart, you go back to the spreadsheet.
With the query visible, you get two practical options:
- Sanity-check it yourself. Read the plain-English question the query represents. If Claude counted "deals" as every open opportunity when you meant closed-won, you'll usually see it in the query's shape and refine the ask.
- Hand it to someone technical for five minutes. A contractor, your IT person, a niece who codes — anyone fluent can scan one query and tell you whether the number is honest. That's a five-minute review, not a BI project.
A dashboard you can verify is a dashboard you'll actually open. That's the whole game.
What to connect first
For most small businesses the answer is simple: your CRM — Salesforce is the one supported connection that isn't a data warehouse. Connect it before anything else.
Here's the honest boundary, though: if your business data lives in QuickBooks, a handful of spreadsheets, and your email, and you don't run a warehouse, this feature isn't for you yet. There's no connector for your stack. Don't buy a Snowflake plan to get a dashboard — that's building a chimney because you liked the smoke. Revisit when either your tools grow native support or your data naturally lands somewhere connectable.
What to actually ask
The gap between "I have a dashboard" and "I use the dashboard" is almost always the questions. Have these ready.
Weekly — movement questions:
- "Which deals moved stages this week, and which ones have been sitting more than 14 days?"
- "Show new leads created by source, this week versus last week."
- "Which accounts haven't been contacted in 30 days?"
These catch problems while they're still cheap — the stalled deal, the lead source that quietly died.
Monthly — pattern questions:
- "Which lead sources produced closed-won deals this quarter versus last quarter?"
- "What's our average time from first contact to close, by month, for the last six months?"
- "Which customers closed 12 months ago and haven't renewed?"
These are the ones nobody asks because assembling the answer takes a day. With a live dashboard, asking takes a minute and the answer stays fresh.
Start with two or three questions, not twenty. The failure mode of self-serve BI is a wall of charts nobody looks at. Three numbers you check every Monday beat thirty you scroll past.
When you've outgrown it
This is a beta feature that regenerates answers from questions. It is not a system of record for your metrics. Move to real BI — or a deliberately built spreadsheet with fixed definitions — when any of these become true:
- More than two or three people need the same numbers every day. At that point definitions need to be fixed and shared, not re-derived per question.
- You need permissions. Salespeople shouldn't see each other's pipelines. If access control matters, you've left consumer-tool territory.
- People start arguing about whose dashboard is right. That's a definitions problem, and it needs one owned source, not a fresh query per conversation.
Until then, a dashboard you can audit by clicking the chart is more than most small businesses have ever had.
One bonus, mostly for the quarterly report
Alongside Dashboards, Anthropic shipped Claude Motion (beta for Team and Enterprise plans): it builds editable animations you can export as MP4. The pragmatic use is the quarterly update — a 30-second animated explanation of where the numbers moved, made from the same data, without touching a video editor or a design tool. If you send stakeholders a quarterly recap nobody reads, a short animation of the same facts is worth testing once.
The Monday habit
Connect the CRM, build the three weekly questions, and put ten minutes on your calendar every Monday to open the dashboard. That's the entire adoption plan. The tool writes the queries; you still have to form the habit — the numbers don't help anyone who doesn't look at them.
If the owner side of this — which numbers to watch, what to check weekly, what triggers action — is the part you keep deferring, it's written down in The AI Agent Owner's Playbook: the permission ladder that matches tasks to trust levels, the first-month inspection habit, and the escalation contract format that tells an agent exactly when to come back to you. $49 CAD, instant download.
Top comments (0)