# Claude Dashboards. | We Call Shotgun > A bank gave me three reasons an AI dashboard can't run their results meeting, so I tested all three: Source: https://wecallshotgun.com/blog/claude-runs-business-dashboards Language: en Author: Toni Dos Santos Published: 2026-11-09 Updated: 2026-11-09 --- Ai # Claude Dashboards. By [Toni Dos Santos](https://wecallshotgun.com/toni-dos-santos) • Nov 9, 2026 • 16 min read • **Summary:** A bank gave me three reasons an AI dashboard can't run their results meeting, so I tested all three: ![](https://substack-post-media.s3.amazonaws.com/public/images/80e5c938-40f4-494e-8fda-95019247b691_1774x887.png)**TLDR** → I pointed Claude Dashboards at an Excel file in my OneDrive. I changed one cell, pressed refresh, and the headline moved. → Before it drew a chart, it asked me four questions about how each number is counted. → Then I had it audit itself. Four Excel mistakes got through without a warning. → The catch: IT has to switch on the Microsoft 365 connector first. I showed managers at a French bank an AI dashboard built from a prompt. They recognised their weekly results meeting in it. Then they gave me three reasons it couldn’t go in. The data isn’t live. It doesn’t connect to our tools. I can’t verify the results. I spent years in that world, from customer support up to corporate banker, and you don’t read out a number you can’t trace. On 8 October, Anthropic shipped Claude Dashboards in beta. I took the bank’s hardest case, an Excel file in Microsoft 365, and ran the three objections against it. Fictional figures. Real file, real connector, real refresh. By the end you’ll have the prompt I ran, five checks, and your first 30 minutes, with or without IT. If the Microsoft 365 connector is already in your Claude list, jump to section 5. Send this to the person who builds the deck for your weekly results meeting. ## 1. The verdict Three out of three, on my test. Live, connected, checkable, on an Excel file in OneDrive. For a team that can get the connector approved, this is the first AI dashboard I’d put in front of a results meeting. ([In May](https://vibeproductmarketing.substack.com/p/build-live-artifact-dashboards-claude) I’d have said no.) Rolling Claude out to a team? Our [Claude training for enterprise teams](https://wecallshotgun.com/claude-training) starts from the work each role already does. ## 2. What it is, and where to switch it on A Claude Dashboard is a web page Claude builds from your prompt. Each time someone opens it, it reruns its queries on your data. Artifacts → Make something new → Dashboard. It’s in beta on Pro, Max, Team and Enterprise ([help centre](https://support.claude.com/en/articles/17454700-get-started-with-claude-dashboards)), and needs code execution on: Settings → Capabilities ([help page](https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude)). On Enterprise an owner switches it on first. ![](https://substack-post-media.s3.amazonaws.com/public/images/3505857e-70eb-4892-b146-17932904797d_1100x270.png)The first screen asks one question: “Use a connector” or “Import a file”. **⚠ Watch out:** “Import a file” takes a copy, and nothing refreshes. Pick the connector. ## 3. The run One prompt, four questions, a dashboard in about ten minutes. My file: four branches, twelve weeks, revenue, new clients, active clients. Sitting in OneDrive. I pasted the prompt from section 5. Two minutes later, questions instead of charts. ![](https://substack-post-media.s3.amazonaws.com/public/images/6b304905-f01b-4810-937a-c983f0344e24_418x370.png)Which week counts as finished, when the sheet already has rows for a week that ends tonight? Is “active” at the end of the week, or at any point in it? Can a client sit in two branches? Is revenue invoiced, booked or collected, with or without VAT? I answered. Seven minutes later the dashboard was up, with two warnings I hadn’t asked for: a week with no rows yet, a branch that stops reporting. ![](https://substack-post-media.s3.amazonaws.com/public/images/8b62d198-2167-435e-ae2d-7d7b47745b16_1265x1180.png)**☑ Tip:** write down how each metric is counted before you open the tool. The questions are coming anyway. ### Objection 1: the data isn’t live **Old:** you paste last week’s export into the chat. The dashboard freezes there. **New:** the dashboard reads the file each time it opens, and the refresh button (top left, next to “Updated”) reads it again. **Why:** the chart is a query on the file, and a copy is a copy. I added €10,000 to Paris for the week of 26 October. Refresh. The headline went from €155,300 to €165,300. ![](https://substack-post-media.s3.amazonaws.com/public/images/622ff6cf-82c0-4740-9790-b4398cc89144_945x530.png)I put the cell back, reopened the link, €155,300. **Missing:** a refresh schedule and alerts (the [help centre](https://support.claude.com/en/articles/17454700-get-started-with-claude-dashboards) only promises a “last refreshed” time per chart). Nothing pings you when a number moves. (In May one of my feeds broke and I didn’t notice for four days.) ### Objection 2: it doesn’t connect to our tools **Old:** export from the CRM, upload, rebuild the chart. Every week. **New:** Claude reads the file where it lives. It also queries Salesforce, and the databases your data team already runs ([help centre](https://support.claude.com/en/articles/17454700-get-started-with-claude-dashboards)). **Why:** the connector is the export, done by the dashboard, every time it opens. The first time the dashboard opened, it asked to use the connector. ![](https://substack-post-media.s3.amazonaws.com/public/images/d802e4df-f2de-4944-afdb-636a1027d400_480x260.png)It only needed to read the file. Click allow. ### Objection 3: I can’t verify the results **Old:** trust the chart, or rebuild the number by hand. **New:** open Sources. The file read, then the calculation, each with a plain-English line on top. **Why:** the dashboard keeps its working, so you check the method once instead of the number every week. ![](https://substack-post-media.s3.amazonaws.com/public/images/0d6fe311-111c-4ed3-a59d-bfb4904d28af_1036x568.png)**★ Remember:** a refresh reruns the calculation exactly as written. If it counts the wrong thing, you get the wrong number, fresher. So I asked Claude to audit its own dashboard. Every number matched my definitions. Then it built broken copies of my file and ran them through the dashboard. Four Excel mistakes got through without a warning. The first one, free: a branch typed in twice for the same week. The headline jumped to €192,900 and nothing on the page flagged it. The other three are in the kit at the end, with the audit prompt. One of them adds €3.7m to the week, and a single character does it. Get the Results Dashboard Kit ## 4. Where it still fails ![](https://substack-post-media.s3.amazonaws.com/public/images/e6728bf9-fada-46c3-b635-e15c5d3f566c_1024x688.png)IT has to set up the Microsoft 365 connector for your company first ([Anthropic, Microsoft 365 connector](https://support.claude.com/en/articles/12542951-enable-and-use-the-microsoft-365-connector)). If it isn’t in your Claude list, that’s the one conversation to have, and the kit has the email. Three more limits. No shared definitions. Two people asking for “churn” can get two different numbers. Mine matched because my prompt forced the questions. Every viewer needs a Claude account and access to the file. A dashboard on a connector can’t be shared with “Anyone with the link” ([Anthropic, sharing](https://support.claude.com/en/articles/9547008-share-artifacts)). The whole file gets read every time. Mine had 48 rows. 50,000 is a test I haven’t run. What your current report still wins: One definition per metric, agreed once, used by everyone. People without a Claude seat can open it. In Power BI, a refresh schedule nobody has to remember. ## 5. Your first dashboard, checked The prompt I ran. It makes Claude list how it’ll count each metric and wait for your yes before it builds. - Open Artifacts → Dashboard. Pick “Use a connector” and choose the one your file lives in. (It isn’t in the list? Section 6 has the other route while IT decides.) - Paste the prompt and fill the fields. - Answer every definition question before Claude builds anything. - Allow the connector when the dashboard asks. ``` I run a weekly results meeting. Build a dashboard for it. Source: {SOURCE_NAME} Table or object: {TABLE_NAME} Metrics: {METRIC_1}, {METRIC_2}, {METRIC_3} Headline: {METRIC_1}, at the top, in large type. Period: the last 8 complete weeks, by week, with the last complete week compared to the week before. Weeks start on {WEEK_START}. Before you write any query: 1. If I left the table blank, list the tables you can see and ask me which one. 2. List how you'll calculate each metric: which field, which filters, which date column. 3. Ask me to confirm or correct each definition. 4. Only then build the charts. Rules: - Every chart uses a live query on the source. No uploaded files. - Under each chart, write the metric's definition in one plain sentence. - If a field name is ambiguous, ask me. Don't guess. ``` `{SOURCE_NAME}`: the connector, as Claude names it (Microsoft 365, Salesforce). `{TABLE_NAME}`: the file and sheet, or the table. Mine read “the Excel file weekly-results.xlsx, sheet weekly_results”. `{METRIC_1} to {METRIC_3}`: the way the meeting says them, headline first. `{WEEK_START}`: the day your reporting week starts. **★ Remember:** the meeting owner keeps the definitions. Claude forgets them between dashboards. Five checks before the meeting. In your first session, run them on the headline only. Sources: open the panel, top right. A connector read means live. “Uploaded” means frozen: ask Claude to rebuild that chart on the connector. The plain-English lines: each one should match what you confirmed. The explanation: in the chat that built it, ask “Explain how this number is calculated, in plain words. What does it include, what does it leave out?” The old way: pull the same number from your usual source, once. If they differ, compare the date range and the filters first. The “Updated” line: top left. Press refresh next to it before the meeting. ## 6. Your first 30 minutes With the connector: Minutes 0 to 5: check code execution is on, open Artifacts → Dashboard, see which connectors you have. Minutes 5 to 10: take three metrics from your last results meeting. Write how each is calculated today, one sentence each. Minutes 10 to 25: run the prompt. Answer the questions with those sentences. Minutes 25 to 30: run the five checks on the headline. Compare it with last week’s slide. Without it, while IT decides: Minutes 0 to 10: the same first two steps. Minutes 10 to 25: export last week’s file, pick “Import a file”, upload it. Change two lines in the prompt: Source becomes “the Excel file I uploaded”, and the first rule becomes “Every chart uses the file I uploaded.” Table stays as your sheet name. Nothing else changes. Minutes 25 to 30: the five checks. Sources will say “Uploaded”, which is why this version won’t refresh. It won’t refresh, but the questions and the five checks still catch what’s wrong in the deck you present this week. I ran this route too. Nine minutes, and Claude still asked its questions first. It also noticed new clients jumping in all four branches the same week, and asked whether the export had changed how they’re counted. Then send IT the request in the kit, so next week’s version reads the live file. > Paid subscribers get the Results Dashboard Kit below: the demo Excel file I tested on, the IT request, a meeting sheet with the metric definitions and the weekly checklist, the full build prompt and the audit prompt. Get the Results Dashboard Kit ## 7. The Results Dashboard Kit Four pieces, in the order you use them, and two files. From the second week, run only the checklist. Start on my file. 48 fictional rows, four branches, twelve weeks, the one I ran every test on. **Want to go deeper?** At [We Call Shotgun](https://wecallshotgun.com/enterprise), we help startups and scale-ups integrate AI into their product and GTM processes. Explore our [AI adoption programs](https://wecallshotgun.com/enterprise) for hands-on workshops and deployment support. ## Keep Reading [Ai Oct 28, 2026 Stop hitting Claude AI usage Limits Why document work burns through usage, and how to build a private converter that gives Claude fewer tokens to read. 8 min read Read →](https://wecallshotgun.com/blog/claude-keeps-asking-for-more-money) [Ai Aug 27, 2026 ChatGPT finally makes AI images that are useful for our brand OpenAI's image 2.0 model is the best for marketing. This week it added transparent backgrounds and stickers. 4 months in, here's what works 14 min read Read →](https://wecallshotgun.com/blog/chatgpt-finally-makes-ai-images-that) [Ai May 8, 2026 Your ChatGPT "Mega Prompts" now make your outputs worse. The 6-part brief replaces your 400-word prompt in ChatGPT 5.5. Stop training AI as an intern. Start briefing it like an agency. 8 min read Read →](https://wecallshotgun.com/blog/your-chatgpt-mega-prompts-now-make) ## Related AI Training Solutions Programs tailored to your role, industry, or function. [Financial Services](https://wecallshotgun.com/ai-training-financial-services) [Healthcare](https://wecallshotgun.com/ai-training-healthcare) [Legal Teams](https://wecallshotgun.com/ai-training-legal) [Operations](https://wecallshotgun.com/ai-training-operations) [Product Teams](https://wecallshotgun.com/ai-training-product) [Marketing](https://wecallshotgun.com/ai-training-marketing) [Sales Teams](https://wecallshotgun.com/ai-training-sales) [Customer Support](https://wecallshotgun.com/ai-training-customer-support) [HR Teams](https://wecallshotgun.com/ai-training-hr) [Data & Analytics](https://wecallshotgun.com/ai-training-data-analytics) [Executives](https://wecallshotgun.com/ai-training-executives) [C-Level](https://wecallshotgun.com/ai-training-c-level) [UK Hub](https://wecallshotgun.com/ai-training-uk) ## Need help with AI Adoption? We take the passenger seat and lead your teams to adopt AI for real work. [Book a Call](https://wecallshotgun.com/#contact)