Julien: Power BI Co-pilot — AI Officer Institute

Starting Point

"I want Claude to design my Power BI dashboard. I'm trying to get Cowork to work directly into Power BI but error messages keep coming up. It's too slow through the browser and doesn't seem able to access it."

Why Browser Automation Doesn't Work Here

Julien was trying to have Claude control Power BI through the browser using Cowork's computer use feature. This approach is too slow and unreliable for Power BI. The interface is complex, with nested menus and drag-and-drop interactions that browser-based AI control can't handle well.

The better approach: use Claude as a thinking partner and instruction generator, not as a remote control for your screen.

Lesson 1

Define the workflow first

Before asking Claude to do anything, talk through the process so Claude understands what you're trying to accomplish. Then let Claude generate the workflow from the conversation.

01

Talk through your steps out loud

Walk through your process in plain language: Where is the source data? How does it get updated? What KPIs matter? What does the end dashboard look like?

Julien already knew the answers. The source data is an Excel file acting as a simple database. He already has a static HTML-based report he likes. The gap is making it dynamic, not redesigning it from scratch.

02

Let Claude generate the workflow

From that conversation, Claude mapped the full process: define the source data, automate the refresh cycle, map KPIs to visuals, connect the pipe so it's no longer static, then validate and iterate.

The important thing: Julien didn't have to write these steps. He talked, Claude organised.

Lesson 2

Create a visual diagram

Once the workflow is mapped in text, ask Claude to generate a visual diagram. This is for the human, not for the AI.

03

Ask Claude to visualise the workflow

Have Claude generate a visual diagram directly in chat. The diagram broke the process into four phases: data prep, connection setup, building the model, and going live.

The visual is a sanity check. If a step looks wrong or out of order, fix it before you start building. The diagram is cheap. Rebuilding a Power BI model is not.

Key Principle

The workflow and the diagram are for the human, not for the AI. You're using Claude to think through and document your process so you can check it, share it, and refer back to it. Claude doesn't need to execute every step. You need to see the steps clearly.

Lesson 3

Look for MCP connectors instead of browser automation

When Claude can't reliably control an app through the browser, the next question is: does a direct connector exist?

04

Check Claude's integrations menu for MCP connectors

MCP (Model Context Protocol) connectors let Claude talk directly to external tools without going through the browser. Cowork currently has 38+ connectors including Slack, Google Drive, Salesforce, Notion, Jira, and Microsoft 365.

Reality check: Power BI is not in the Cowork connector directory today. This may change, but as of April 2026 it's not there.

05

Explore third-party options

Third-party MCP servers exist for Power BI, including Microsoft's Power BI Modeling MCP Server (a VS Code extension) and CData Connect AI. These work with Claude Desktop's developer mode and Claude Code, not with Cowork's one-click connector system. They require manual JSON config and technical setup.

06

The fallback: Claude generates instructions, you execute

If no direct connector exists, have Claude generate detailed step-by-step instructions you implement manually in Power BI. Claude writes the DAX measures, designs the data model, and maps the KPIs. You paste and build.

This is still an 80% time saver. Claude does the thinking. You do the clicking.

The Hierarchy of AI Assistance

Best case: Direct MCP connector. Claude talks to the tool natively. Minimal human clicking.

Middle case: Third-party MCP server. Works but requires developer-level setup. Best for technical users.

Fallback (most common today): Claude generates step-by-step instructions and code. You execute in the tool. Still saves 80% of the work.

Lesson 4

Store workflows in a catalog

Once you've built a workflow, don't let it die in a chat thread. Save it, publish it, share it with your team.

07

Save workflows to a shared folder and publish

Save workflows to a folder structure, publish them to the web via GitHub and Vercel, and share them across the team. Claude effectively replaces a CMS for maintaining this catalog.

The next person who hits the same problem can pull up the documented workflow instead of starting from scratch.

08

Think about access governance early

Who can access and edit shared workflows? The current approach uses folder-level permissions within Claude Teams shared projects. Team members can publish to specific folders but can't touch other parts of the catalog.

Define who can publish what, and where — even if it's informal at first.

Prompts You Can Use

Prompt 1 — Map the workflow

I want to move my company's data from a static Excel file into a live Power BI dashboard. Walk me through the process step by step. The source data is [describe your Excel file: what tabs, what columns, how often it's updated]. I already have a report layout I like that shows [describe your KPIs]. I need the dashboard to auto-refresh [daily/weekly/monthly]. Ask me clarifying questions before generating the workflow.

Prompt 2 — Generate a visual diagram

Now create a visual diagram of this workflow. Show the phases and steps so I can check whether the order makes sense before I start building.

Use this after Prompt 1. The diagram is your sanity check. Review it, flag anything that looks wrong, and iterate before moving on.

Prompt 3 — Structure the Excel for Power BI

Review this Excel file and tell me what I need to clean up before connecting it to Power BI. Check for: merged cells, inconsistent data types, missing headers, duplicate columns, and anything else that would cause problems during import. Give me a checklist of fixes.

Upload your Excel file with this prompt. Claude will flag structural issues before they become Power BI errors.

Prompt 4 — Write DAX measures

I'm building a Power BI dashboard. My data model has these tables and columns: [list them]. I need DAX measures for the following KPIs: [list your KPIs, e.g., 'total revenue by month', 'year-over-year growth rate', 'average deal size by client category']. Write the DAX for each measure and explain where to paste it in Power BI.

This is where the 80% time savings happens. Claude writes the DAX. You paste it into Power BI's measure editor.

Prompt 5 — Design the dashboard layout

I have these KPIs ready in my Power BI model: [list your measures]. Design a dashboard layout for me. Tell me which visual type to use for each KPI (bar chart, card, line chart, table, etc.), how to arrange them on the canvas, and what filters or slicers to add. I want the dashboard to be clean and scannable for [describe your audience].

Claude can't drag visuals onto the Power BI canvas, but it can design the layout and tell you exactly what to build.

Prompt 6 — Set up auto-refresh

Walk me through setting up scheduled refresh for my Power BI dashboard. My Excel source file is stored on [OneDrive/SharePoint/local]. I want it to refresh [daily/weekly/monthly]. Give me the exact steps in Power BI Service to configure this, including any gateway setup if needed.

Prompt 7 — Validate the dashboard

I've built my Power BI dashboard. Here are the numbers it's showing: [paste key figures]. Here are the numbers from my original static report: [paste the same figures from your old report]. Compare them and tell me if anything doesn't match. If there are discrepancies, help me figure out where the data is diverging.

Prompt 8 — Document the workflow for the team

Create a step-by-step guide documenting everything we just built: the data source, the refresh setup, the DAX measures, and the dashboard layout. Format it so someone on my team who wasn't part of this process can maintain and update the dashboard without my help. Include a troubleshooting section for common issues.

This becomes the catalog entry. Save the output and publish it to your team's shared folder.

The Four Moves

When Claude can't directly control a complex tool, this is the playbook.

01

Define the workflow with Claude

Talk through the process. Let Claude map the steps. Generate a visual diagram to sanity-check it.

02

Check for MCP connectors

Look in Claude's integrations for a direct connector. If one exists, use it. If not, check for third-party MCP servers.

03

Fall back to Claude-generated instructions

No connector? Have Claude write the step-by-step guide, the DAX measures, the data model. You execute in Power BI. Still saves 80% of the work.

04

Save the workflow to your catalog

Don't let it die in a chat. Save it, publish it, share it with the team.