As a self-employed person, I've been thinking about car leasing for years. The problem isn't finding quotes—it's comparing them. Different contract lengths, residual values, mileage allowances, benefit values, deduction possibilities. Every time I've tried to build a proper comparison, it's taken days and I've still felt unsure if I missed something.
Last week I tested Claude in Excel. In fifteen minutes I had a complete model: eight tabs with leasing options for three car models, three contract lengths, sensitivity analysis for mileage and residual value, tax effects for sole proprietorship vs. limited company, and a summary dashboard showing total monthly cost for each scenario.
That's not an improvement. It's a different category.
What's Different
Claude understands the structure of your workbook. Every tab, every formula, every cell reference. Ask it to explain how the model works and it traces formula relationships down to the cell level. Ask it to update an assumption and it modifies cells without breaking dependencies.
The interesting thing about the leasing model was that Claude suggested analyses I hadn't asked for. When I mentioned I was self-employed, it automatically added a tab for tax effects and a comparison between private leasing and business leasing. It asked follow-up questions about my approximate marginal tax rate and annual mileage.
Full traceability makes it actually useful. Every action is logged. If you want to understand why a cell shows a certain value, you can trace the logic backward. That's the difference between an experiment and something you actually dare to base decisions on.
Why Data Is the New Moat
We've spent years discussing who has the smartest model. That discussion is becoming irrelevant.
Anthropic has quietly built licensed partnerships with the data providers the financial world actually runs on: LSEG for real-time market data, Moody's for credit ratings, S&P Capital IQ for financial data, Morningstar and Pitchbook for analysis.
A generic language model can write a SUMIF formula. Claude can fetch this morning's pricing from the London Stock Exchange, cross-reference it against Moody's credit ratings, and update your analysis. In a single flow.
The quality of AI output depends entirely on input quality. Exclusive data therefore becomes an enormous advantage.
The Peculiar Partnership
Microsoft and Anthropic have a partnership worth 30 billion dollars. Microsoft hosts Claude on Azure. At the same time, Claude competes directly with Microsoft's own Copilot for Excel.
There's a product difference that matters: local file handling. Copilot requires files to be saved on OneDrive with autosave. Every change syncs directly to the cloud.
Many hate this. They want control over when work is saved and the ability to experiment without permanent changes. Claude works with local files. For me as a self-employed person with sensitive calculations, that's valuable.
What Works and What Doesn't
Let's be honest about the limitations.
The positives:
- Speed is remarkable. My leasing model took fifteen minutes. By hand it would have taken at least a full day.
- Multi-tab architectures are handled excellently. Eight tabs with cross-dependencies worked without issues.
- The model is proactive. Tax effects and opportunity cost analysis it suggested itself.
- The price is accessible: 20 dollars a month.
The challenges:
- Specific figures (exact benefit values, current interest rates) I needed to input manually.
- Charts are functional, not pretty. For a presentation you'll need to do the formatting work yourself.
- For really large models you may need to restart the chat due to memory limitations.
What It Means
Norway's sovereign wealth fund estimates they've saved 213,000 work hours by using Claude in Excel. Those are numbers that are hard to ignore.
But the most important thing isn't the time savings. It's what happens to the value chain of analytical work.
Calculation becomes cheap. Building a model, running sensitivity analysis, comparing scenarios—that has historically required days. Now it takes minutes.
Judgment becomes expensive. Knowing which assumptions are reasonable, which scenarios are relevant, which conclusions hold—AI cannot deliver that.
My leasing model gave me answers, but I still needed to determine whether the assumptions were reasonable. How likely is it that I drive exactly that mileage? How do I value flexibility vs. lower monthly cost? Those are judgment questions no model can answer.
Actions
1. Test on a real problem. Not a toy example. Take something you actually need to calculate—an investment decision, a pricing model, a leasing comparison. 20 dollars a month isn't worth discussing.
2. Focus on judgment. Automate the calculation. Spend time asking the right questions and validating the answers. That's where the value is.
3. Be critical of assumptions. AI builds models quickly, but it doesn't know if your assumptions are reasonable. That's still your job.
4. Use it for decisions you've postponed. Most of us have financial questions we've avoided because the analysis felt too time-consuming. That excuse no longer exists.
Conclusion
The spreadsheet is the business world's nervous system. From board decisions valued in billions to the self-employed person's leasing choice—everything flows through its cells.
Claude has made analytical work dramatically cheaper. That doesn't just change speed—it changes what's worth paying for.
The question isn't whether AI is good enough. The question is which decisions you've postponed because the analysis felt too expensive—and what happens when that cost disappears.