Claude in Excel: The Complete Tutorial for Finance Professionals

As a technology consultant, I spend a lot of time inside Excel. Business cases, scenario analyses, cost-benefit models with sensitivity tables — the kind of workbooks with 15 tabs all referencing each other. It's not glamorous, but it's where most consulting deliverables live. So when Anthropic released Claude in Excel, I didn't just try it. I stress-tested it against the exact workflows I use on real client engagements.

This is what I found. The good, the bad, and the one time it confidently built a circular reference that took me 20 minutes to untangle.

What Claude in Excel Actually Is (and Isn't)

Claude in Excel is Anthropic's add-in for Microsoft 365 that embeds Claude directly into your spreadsheet sidebar. Unlike Copilot — which I've used since its launch and found limited past basic tasks — Claude reads your entire workbook context. Every tab, every named range, every cell relationship.

In practical terms: when you ask Claude to calculate operating cash flow, it doesn't just look at the cell you're in. It reads your Income Statement tab, finds net income, cross-references the Balance Sheet for depreciation, and builds the formula with correct cross-tab references. It then cites the exact cells it used, so you can verify every input.

That's a meaningful difference from Copilot, which I've found loses context once you're working across more than two or three tabs. I've written about why context matters for AI tools — this is a perfect example of it in practice.

My Setup

Before we start: I'm running this on Microsoft 365 desktop (Version 2602, Windows 11) with a Claude Pro subscription. The add-in appeared in the Office store in late January 2026. Setup took about 3 minutes:

1. Insert > Add-ins — search "Claude for Excel"
2. Install and authenticate with your Claude account
3. The sidebar appears on the right — that's your interface

One thing I wish I'd known upfront: Claude can read your entire workbook, but it can only write to one tab at a time. If you need formulas across multiple tabs, you'll make separate requests. Not a dealbreaker, but it changes how you plan your workflow.

Building a Business Case Model: What I Actually Did

I didn't build a toy example for this tutorial. I rebuilt a simplified version of a business case I'd previously delivered on a technology transformation engagement — a cost-benefit model with projected costs, expected savings, and scenario analysis across different adoption rates. The original took me about 6 hours to build and validate manually. I wanted to see how much Claude could compress that.

Structuring the Workbook (Still Manual)

I set up three tabs myself: Cost Model, Benefits Case, and Scenario Analysis. Row headers, time periods (Q1-Q4 2026), section groupings, assumption callouts. Claude works best when the skeleton exists and it fills in the logic — asking it to create the entire structure from scratch produced a layout that didn't match how I actually organise business cases for clients.

Lesson learned early: give Claude structure, ask it for logic. Not the other way around.

Cost Projections and Scaling Formulas

My first prompt:

"Build implementation cost formulas in the Cost Model tab. Q1 baseline spend is 150,000 for licensing and integration. Apply 15% quarterly reduction as onboarding costs taper. Add a fixed operational cost of 25,000 per quarter from Q2 onwards."

Claude nailed this. It placed formulas in the correct cells, used relative references so they'd copy cleanly across quarters, and separated one-off implementation costs from ongoing operational costs — exactly how I'd structure it for a client.

What I appreciated: it didn't just write =D5*0.85. It wrote =D5*(1-$B$3) — referencing a cost reduction rate in a dedicated assumptions cell I could change later. It inferred that I'd want dynamic assumptions even though I didn't ask for one. That's the difference between a quick calculation and a model you can actually present to stakeholders.

Cross-Tab References (Where It Pulls Ahead)

This is the test that matters. I asked:

"In the Scenario Analysis tab, calculate net benefit by pulling total costs from the Cost Model and total savings from the Benefits Case, then show cumulative net position by quarter."

Claude built ='Cost Model'!D18 and ='Benefits Case'!D12 without me specifying cell addresses. It found the right totals by reading the row labels across tabs and matching them to my request. Then it added a running cumulative sum to show the breakeven point.

I verified every reference manually. All correct. On a client engagement, I'd still verify — but the time saving is significant. What used to take 30-40 minutes of tab-switching and cell-hunting took about 90 seconds.

Sensitivity Tables

"Build a sensitivity table showing net benefit at adoption rates of 60%, 75%, 85%, and 95%, crossed with implementation timelines of 3, 6, 9, and 12 months."

Claude generated a two-variable data table linked to the assumption cells. Clean, functional, correctly structured. This is the kind of deliverable that makes a business case persuasive — showing decision-makers the range of outcomes, not just the base case. I've spent hours building these manually on past engagements, and Claude handled it in one prompt.

Where It Broke

I'm being honest about this because I think it matters more than the success stories.

When I asked Claude to add a phased investment schedule where savings in one quarter fund the next quarter's implementation — a common pattern in transformation business cases — it created a circular reference. Savings depended on implementation progress, which depended on available budget, which depended on savings. Claude built each formula correctly in isolation, but it didn't flag the circularity until I noticed the #REF errors.

I had to manually break the loop by hardcoding the investment schedule and using it as a fixed input rather than a dynamic feedback loop. Claude couldn't help with that fix — it kept suggesting formula changes that recreated the same circular dependency from a different angle.

This is important context. Claude in Excel is excellent at linear formula chains and cross-tab references. It struggles with circular dependencies — which come up more often than you'd think in business cases where costs and benefits are interdependent. When your model has feedback loops, build those sections yourself and let Claude handle the linear calculations around them.

Honest Comparison: Claude in Excel vs Copilot

I've used both extensively. Here's where I've landed as of March 2026:

Capability: Multi-tab awareness | Claude in Excel: Reads full workbook context | Copilot in Excel: Loses context past 2-3 tabs

Capability: Nested formula generation | Claude in Excel: Handles 4-5 levels reliably | Copilot in Excel: Unreliable past 2-3 levels

Capability: Cell-level citations | Claude in Excel: Cites every source cell | Copilot in Excel: Not available

Capability: Cross-tab references | Claude in Excel: Infers correct cells from labels | Copilot in Excel: Needs manual cell addresses

Capability: Circular dependency handling | Claude in Excel: Does not detect or resolve | Copilot in Excel: Does not detect or resolve

Capability: Scenario modelling | Claude in Excel: Creates linked data tables | Copilot in Excel: Basic tables only

Capability: Speed to install | Claude in Excel: 3 minutes | Copilot in Excel: Pre-installed with 365

Capability: Price | Claude in Excel: Claude Pro ($20/mo) | Copilot in Excel: Included in Copilot 365

Neither tool handles VBA, macros, or pivot tables yet. Both are formula-and-analysis tools at this stage.

What I'd Do Differently Next Time

After this build, I'd change my approach on the next client engagement:

1. Create a dedicated Assumptions tab first — before asking Claude to build any formulas. It's much better at referencing existing assumption cells than creating them on the fly. On consulting deliverables, a clean assumptions tab also makes the model easier for clients to review and challenge.
2. Map your circular dependencies before you start — if your business case has feedback loops (and most transformation cases do), build those sections manually and let Claude handle the linear chains around them.
3. Verify cross-tab references before you present — Claude was accurate in my testing, but a wrong cell reference in a client deliverable has real consequences. I verified every one. You should too.
4. Use it for debugging existing models — pasting a broken formula into Claude's sidebar and asking "why doesn't this work?" turned out to be one of the most useful features. I had a legacy workbook from a previous engagement with a broken SUMPRODUCT — Claude identified the issue in seconds by reading the surrounding cells. That alone saves hours on inherited models.

Who This Is For

Claude in Excel is worth the Claude Pro subscription if you:

• Build business cases, cost-benefit analyses, or financial models as part of your consulting or strategy work
• Spend significant time debugging formula chains across complex multi-tab workbooks
• Need sensitivity tables and scenario analyses for stakeholder presentations
• Want AI that understands your full workbook — not just the active cell

If your spreadsheet work is mostly single-tab data entry or basic charts, Copilot does the job. Claude in Excel is built for the kind of complexity that consultants, analysts, and project leads deal with on every engagement.

What's Coming Next

I'm building follow-up tutorials on specific use cases: technology transformation business cases, budget variance analysis, and project portfolio tracking. Each one will be a real model built from a real use case — not a contrived example.

Subscribe to the newsletter and I'll send them as they're published.

Shea Campbell is a technology consultant and AI engineer who builds intelligent automation for real workflows. [Subscribe to Ops & Insights](/newsletter) for practical AI tutorials every week.