Claude’s integration into Excel allows complex spreadsheets to be analyzed and edited in real time with AI. The native integration offers a pretty smart experience.
This was an integration awaited for several months. Since mid-January, Anthropic’s artificial intelligence can be integrated as a plugin into Microsoft Excel. The only condition? Have an Anthropic subscription (Pro or Max) and a Microsoft license. AI does not replace Copilot but works differently. Where Copilot has a basic use as an assistant in Excel, Claude behaves almost like a full-fledged agent. It understands the entire structure of a file and can edit multiple tabs at the same time. In short, the integration is much more thorough. Anthropic positions its tool as essential for financial services.
How to install Claude in Excel?
Installing Claude in Excel couldn’t be simpler. Simply go to the Microsoft marketplace and install the “Claude by Anthropic in Excel” add-on. Once installed, Claude integrates into the left side panel of Microsoft Excel instead of Copilot, even if the two AIs can be used simultaneously. You must then log in to your Claude account. The configuration is then completed.
Claude’s integration into Excel is quite similar to the native application. There is a model selector to choose between Sonnet 4.5 and Opus 4.5, a chat and several options: web search, adding a file or a photo to the conversation and exchange history. Two AI action modes are available: ask before acting and act without asking. Concretely, with the first mode, Claude will ask your permission before modifying the file, and with the second, he will modify in total autonomy. It is of course possible to revert to a previous version using Excel versioning.
Thanks to its partnership with Microsoft, Anthropic was able to fully integrate Claude into the Excel code. The AI is capable not only of understanding the overall context of a file (only XLSX and XLSM formats are supported) but also of acting on several tabs at the same time to retrieve formula references. You can ask questions and get answers with quoting source cells, update formulas while preserving dependencies, create graphs… AI acts for you. To illustrate these capabilities, we tested four very concrete use cases: analyzing the structure of a complete file at a glance and generating documentation, quickly debugging a complex formula, cleaning raw data and quickly simulating a scenario.
1. Document a complex file
For this first example, let’s take for example the financial statements of the French State from 2006 to 2024. A complex file which clearly requires documentation. We simply ask the AI: “Analyze this file and generate complete documentation: purpose of each tab, dependencies between sheets, key formulas used.”
Claude then executes and generates precise documentation in a new tab with the overview, the different elements of each tab and the different dependencies and formulas identified. Interestingly, to remember past conversations, Claude uses an Excel tab with summaries of your exchanges.
2. Debug a formula
Who has never racked their brains over an Excel formula with a #VALUE error! or #NAME! ? Here again, Claude in Excel can work miracles. Simply ask the AI to understand the formula, break it down and identify the potential error(s) and then suggest a fix. Example in our State accounts file, once the problematic cells have been selected, we ask the AI: “these interdependent formulas return an error. Identifies the error, proposes an appropriate correction and applies it.”
The AI thus identified that the formulas H5 to H7 mistakenly referenced an empty column and texts. To reconstruct the missing data, Claude then estimated their value from the total fixed assets (H8) by category estimated by interpolation between 2018 and 2020. More than a simple error correction, the AI managed to extrapolate the data when it understood that the data was not present in the table. Interesting.
3. Quickly clean raw data
Even more interesting, it is possible to ask Claude to quickly clean a complex and/or poorly documented file, as is often the case with open data files. We take for example the file of the results of municipal elections in the second round in municipalities with less than 1000 inhabitants in France. Many data are poorly structured (mixed text/number types, empty column, mixed department codes, etc.) We ask the AI to clean the file and classify by decreasing participation rate.
Claude complies, once again, without the slightest obstacle. The data is cleaned and filtered perfectly and the file becomes fully usable.
4. Simulate a hypothesis
For this last test, we are using the file of financial statements of the French State from 2006 to 2024. The goal will be to have Claude simulate hypotheses based on the data. We will therefore ask: “Based on historical developments and the net situation, project 3 scenarios for 2030: continuation of the current trend, reduction of the deficit by 30% and worsening of the deficit by 20%. Create a simulation tab with the projections side by side and a comparative graph.”
Claude then analyzes the entire report and generates a complete tab with historical data, comparative projections 2024-2030, graph and summary. Asked about the difference between his forecast and that actually observed in 2025, Claude, after a web search, concludes that the deficit observed (-€124.7 billion) is better than expected because it is based on tax increases. “This is encouraging, but the 0.6 point reduction in the deficit provided for by the finance law is based almost exclusively on tax increases rather than on savings,” says the IA.
Pay attention to your quota
Claude in Excel is a extremely effective tool. Automatic documentation, intelligent debugging, raw data cleaning, forward-looking simulation: the AI executed each task without a single error, systematically understanding the overall file context and dependencies between tabs. We are far from a simple chatbot grafted onto a spreadsheet: Claude behaves like a real analyst. Used correctly, the time savings can be considerable.
A major downside, however: the consumption of tokens. In less than an hour of testing with a Claude Pro account, we consumed 80% of our quota. Each interaction with a large file eats up a significant amount of tokens. For intensive daily use, the Pro subscription (20 dollars per month) may prove insufficient and it will probably be necessary to consider the Max offer (100 dollars per month minimum). A parameter to take into account before deploying the tool on a team scale.
