How I Used Claude AI for Excel to Analyse 1,200 Orders — And Found $520K in Missed Revenue

I spent hours building a multi-variable revenue breakdown. Charts, tables, cross-referenced data by product, coupon code, and channel. I had everything. Except the one variable that mattered most. Claude caught it — and it wasn’t even what I asked for.

What Is Claude AI for Excel?

Claude AI for Excel is an official plugin from Anthropic that brings the Claude model directly into your Microsoft Excel ribbon. Instead of exporting data to another tool or writing formulas you half-understand — you open a chat panel inside Excel and tell Claude what you want in plain English.

For marketing teams doing analytical work, this matters for three specific reasons:

• You can clean raw exports from ad platforms, CRMs, and ecommerce tools without touching a formula
• You can request multi-variable breakdowns — revenue by product by coupon code by channel — in a single prompt
• You can ask open-ended questions and surface findings you weren’t specifically looking for

It is not a replacement for analytical thinking. But in the right hands, it compresses hours of mechanical work into minutes — which is what the rest of this article demonstrates.

---

Step 1 — How to Install Claude AI for Excel

In Excel, go to the Home tab → Add-ins → Office Add-ins marketplace and search for “Claude for Excel.” Install it, authorise the connection to your Claude account, and you’ll see the Claude icon appear in your ribbon. Click it to open the chat panel on the right side of your spreadsheet.

Make a backup copy of your spreadsheet before doing anything else. Claude has permission to write directly to your Excel file. Always work on a copy — not the original — until you’ve built confidence in the workflow.

---

Step 2 — How to Clean a Messy Marketing Dataset with Claude AI

The scenario: a marketing analyst needs to deliver an executive summary — findings, charts, recommendations — within 24 hours. The dataset is a 1,200-row order export: twelve columns, products, order status, marketing channels, coupon codes, and revenue. Real data. Messy. No prior cleaning.

Most tutorials skip this and start with clean data. That’s not useful. So the first test: can Claude handle the file exactly as it is?

Select the entire dataset, open the Claude panel, and start with a diagnostic:

Can you explain this spreadsheet to me.

Claude walks through the columns and structure — useful when you’ve inherited someone else’s file. Then the real test:

Can you identify issues with this spreadsheet and clean them up.

Claude runs through the sheet and makes changes. Rather than accepting them blindly, ask Claude to make every change visible:

For all the changes you’ve made, can you highlight them in a colour so I can review.

Claude highlights the changes in different colors – blue cells show formula updates, yellow cells are autofills. Filtering the Coupon Code column by yellow reveals every blank has been filled with “NONE” — clean, consistent, reviewable. Something most analysts do manually, done through a chat prompt in about ten seconds.

Claude’s autofill logic is based on pattern recognition. In this case filling blanks with “NONE” was correct — but always filter the highlighted cells and verify the logic before moving on. Do a quick row count to confirm no data was lost. Treat it like reviewing a teammate’s work: the heavy lifting is done, the sign-off is yours.

---

Step 3 — How to Build a Multi-Variable Revenue Breakdown

Data clean and verified. Now the actual analysis. Ask Claude to create four summary tables on a new sheet:

In a separate sheet, can you summarise sales by product, by order status, by coupon code, and by referral source.

Four tables appear. There’s a small formatting error in the first — not surprising, the prompt wasn’t specific about layout. Fix it, then run an audit before building on top of the tables:

Before we go any further, can you audit all four tables and confirm the numbers are correct.

Never build analysis on top of tables you haven’t verified. Errors compound upward — a miscounted row in a summary table leads to a wrong chart leads to a wrong executive finding. This prompt takes 10 seconds and saves you from a very awkward room.

Then the three-variable combined breakdown — and this is where the prompt evolution matters:

FIRST ATTEMPT

Can you create a single table showing sales for each product, broken down by coupon code and referral source together.

It produced separate tables for product revenue by coupon code and product revenue by referral source

SECOND ATTEMPT : WHAT ACTUALLY WORKED

I need all three variables — product, coupon code, and referral source — combined in one table, not separate tables. I’m trying to understand which combinations of coupon code and channel are driving revenue for each product specifically. Can you rebuild this as a single unified table.

The key was explaining what I was trying to understand, not just what format I wanted. Context about your intent produces better structure than format instructions alone — this is the most transferable prompting lesson in the whole workflow.

Can you now show the same breakdown as a percentage of overall product revenue, rather than absolute dollar values.

Percentage errors are harder to spot than absolute number errors — they look plausible even when wrong. Always manually verify two or three rows against the raw totals. Pick the largest and smallest values. This takes 90 seconds and gives you full confidence in the table.

---

Step 4 — How to Generate Charts with Claude AI for Excel

With clean data and verified tables in place, charts take one prompt:

Can you create a revenue chart for each product broken down by coupon code and referral source.

Charts appear in seconds. Numbers check out — but they’re in dollar values. For an executive presentation, percentage share is cleaner:

Can you rebuild these as percentage revenue share rather than absolute values. One chart showing coupon code performance across all products. One showing referral source performance across all products.

Charts and pivot tables are the area where Claude requires the most back-and-forth. Expect to refine once or twice — especially around axis labels, legend formatting, and colour schemes. The underlying data is usually right; the presentation layer needs more iteration than cleaning or tables. Budget time for this.

---

Step 5 — How Claude Found the $520K Opportunity I Completely Missed

Clean data. Four tables. Multi-variable breakdown. Percentage charts. Everything needed for an executive summary. But before writing it up, I asked one more question — framed deliberately wide:

Can you identify executive-level findings from this dataset — and focus specifically on items with high revenue impact or high variance. I want to surface both the top performers and the outliers.

The first thing Claude came back with was not about products. Not about coupon codes. Not about referral sources.

It was about cancellations.

In all my analysis — across three variables, two chart types, four summary tables — I had never once filtered this dataset by order status. That dimension never came up. According to Claude, cancelled orders represented the single biggest revenue impact in the entire dataset. Bigger than any product. Bigger than any coupon code combination. $520K sitting in a column I had simply never looked at.

This isn’t just about Claude running fast. It ran a parallel check across dimensions you weren’t monitoring — and surfaced something you would have walked into a boardroom without knowing. Open-ended prompts with variance framing catch what narrow prompts miss. The prompt you write defines the ceiling of the analysis. Write it wide.

Claude flagged that cancellations are significant — but it cannot tell you why they’re happening. That answer might live in your customer service tickets, your returns policy, your fulfilment partner SLA, or a seasonal pattern. Claude finds the signal. Interpreting that signal requires you.

---

Claude AI for Excel: An Honest Strengths and Weaknesses Assessment

After running this plugin through a full end-to-end marketing analysis workflow, here is where it earns its place — and where it still needs you.

Final Verdict: Should Marketing Teams Use Claude AI for Excel?

The honest answer is yes — with a clear understanding of what it does and doesn’t do.

For marketing teams that regularly export data from ad platforms, ecommerce tools, or CRMs into Excel, the time compression on data cleaning and table generation alone justifies the install. What took the better part of an afternoon now takes under ten minutes.

But the more significant value isn’t speed. It’s the expanded surface area of the analysis. When you run a narrow analysis, you find what you were looking for. When you give Claude an open brief to surface variance and high-impact findings, you find what you weren’t looking for — and that second category often turns out to be the most important.

The $520K cancellation finding wasn’t in any of my tables or charts. It was in a column I hadn’t looked at — and Claude found it because I asked the right final question. The plugin is a force multiplier for analysts who already know what questions to ask. Used well, by the right person, with the right prompting discipline — it is one of the most practically useful AI tools available in a marketing analyst’s stack right now.

Claude AI for Excel is not a replacement for analytical thinking. It is a compression of the mechanical work that sits between thinking and insight — and occasionally, it will find the thing your thinking missed. That combination is worth taking seriously.