· Saul Mateos · Claude and AI Tools  · 6 min read

AI Dashboards for Finance: CSV to Dashboard in 15 Minutes

AI dashboards for finance with Claude Code: the prompts that turn one CSV into a dashboard you'd show a CEO, plus the verification loop and a sample file.

Here’s how I build AI dashboards for finance. My CEO pings me asking for a quick view on churn versus new logo growth. Normal request. I opened Claude Code, dropped two CSVs, typed one sentence, and had a working chart with commentary in 4 minutes. Six months ago that’s a 45-minute exercise minimum.

This is the session I run as Claude Code for the Finance Operator: install it, then use it on real work. Not a demo. Your actual data. If you don’t have a file handy, use the sample SaaS metrics file: 12 months of MRR, churn rate, new customers, headcount, the usual.

Download the sample SaaS metrics file (CSV)

Getting started with Claude Code

# 1. Install Node.js from nodejs.org (LTS version)
node --version  # verify it's installed

# 2. Install Claude Code
npm install -g @anthropic-ai/claude-code

# 3. Launch it
claude

# 4. Authenticate (API key or Claude Max subscription)
# 5. Navigate to a folder with your data
# 6. Start typing in English

“This is it. You type English. It writes code and runs it. No IDE required. No plugins. Just a conversation in the terminal that can actually do things on your machine.”

Read the file sample-saas-metrics.csv. Tell me: what are the 3 most important trends a CFO should flag for a CEO update? Be specific with numbers.

Notice what I did. I didn’t say “analyze this data.” I gave it a role (CFO), an audience (CEO), and a constraint (3 trends, specific numbers). That’s the difference between mediocre output and useful output.

The 5-part prompt framework

Every good prompt has some combination of:

  1. Role: Who is AI acting as? (“As a CFO preparing for a CEO update…“)
  2. Task: What specifically do you want? (“Identify the top 3 trends…“)
  3. Context: What does AI need to know? (Your data, audience, constraints)
  4. Format: How should the output look? (“Table” / “3-sentence summary” / “HTML dashboard”)
  5. Constraints: What are the boundaries? (“Under 200 words. Only use data from this file.“)

You don’t need all five every time. But the more you give, the better the output.

Step 2: Build AI dashboards for finance from one prompt

Using the data from sample-saas-metrics.csv, build an interactive HTML dashboard with:
1. A line chart showing MRR trend over 12 months
2. A bar chart showing monthly churn rate with a target line at 3%
3. A summary card row at the top showing current MRR, current churn rate, headcount, and net revenue retention
4. Use a clean, professional color scheme. Dark sidebar, white content area.
Make it look like something I'd actually show a CEO.

It’s writing HTML, CSS, JavaScript right now. Chart library, layout, data binding. All from one prompt. You typed English. It’s building software.

That took about 2 minutes. Is it perfect? No. But it’s 85% there. Now watch what happens when we iterate.

Step 3: Iterate, don’t start over

Make the MRR chart a waterfall showing month-over-month change instead of a line. Green for growth months, red for decline.
Add a trend line to the churn chart. And below the charts, add a 3-sentence executive summary of what this data means for the business.

See the pattern? I’m not starting over. I’m refining. “Make this a waterfall.” “Add a trend line.” Each prompt builds on the last. That’s how you talk to this tool. Don’t start over. Refine.

Step 4: The verification loop

Show me the raw data you used for the summary cards. Display the actual values from the CSV next to what's shown on the dashboard so I can verify accuracy.

AI generates. You verify. Every time. Especially numbers.

AI will confidently show you $8.2M when the actual figure is $7.1M. It happens regularly. Never send AI-generated financials without checking the source data.

The 15-minute rule

  • 0-5 min: Try AI with a clear, specific prompt
  • 5-10 min: Refine the prompt if the initial result is close
  • 10-15 min: One more attempt with a different approach
  • At 15 min: Decision time. Working? Keep going. Stuck? Bail to Excel. No guilt.

This rule saves you from 3-hour debugging spirals. AI is a tool, not a religion.

Then try it on your own work

What’s a task you do every week that takes too long? Something repetitive. Something you dread.

Report writing and commentary

I'm going to paste [monthly financial data / variance report / key metrics]. Write a 3-paragraph executive summary for my CEO. Focus on: what changed, why it matters, what we're doing about it. Be specific with numbers. Keep it under 200 words.

Feed it the source data. Give it the audience and the format. Then edit the output. You’re going from blank page to 80% draft in 30 seconds.

Data cleanup

I'm going to paste a messy spreadsheet. Clean it up: standardize the column headers, flag any rows with missing data, remove duplicates, and give me a summary of what you fixed.

This is the boring work that eats Tuesday afternoons. AI loves boring work. Let it clean, you review.

Email and memo drafting

Draft an email to [audience] about [topic]. Tone: direct but professional. Key points: [list 3 things]. Keep it under 150 words. I'll edit from there.

Don’t ask it to write the final version. Ask for a first draft. You edit. That’s the workflow.

Comparison and analysis

I'm going to give you two [documents / datasets / proposals]. Compare them and tell me: what's different, what's missing from each, and which one is stronger on [specific criteria]. Format as a table.

Any time you’re comparing two things side by side, feed them both in. AI is great at structured comparison.

Your 4-week roadmap

Week 1: Use Claude Code for one task per day. 15 minutes max each. Low stakes. Get comfortable with the conversation pattern.

Week 2: Pick your most annoying recurring report. Try generating it. See how close AI gets.

Week 3: Build your first reusable prompt or workflow. Something you can run again next month without starting from scratch.

Week 4: Evaluate. What’s saving time? What’s not? Kill anything that doesn’t save 2+ hours per week. Double down on what works.

Everything here used the terminal (Claude Code). If you want to give this to team members who aren’t going to live in a command line, check out Claude Cowork. Same AI, desktop app, point it at a folder. That’s how you scale it.

Download the sample SaaS metrics file (CSV)

Share:
Back to Blog

Related Posts

View All Posts »