AI Excel: How Artificial Intelligence Is Changing Spreadsheet Work

The three waves of AI in Excel

The first wave was formula help. Tools like FormulaBot and early ChatGPT plugins could write a VLOOKUP or explain a nested IF statement. Useful, but incremental. You still had to know what formula you needed and where to put it.

The second wave was data analysis. Microsoft Copilot and Google Gemini brought natural-language queries to spreadsheet data: "What were sales by region last quarter?" These tools are good at summarizing existing data but struggle to build anything new.

The third wave, the one we're in now, is AI agents that build. Instead of answering questions about data you've already entered, these tools construct entire financial models from a plain-language description. Three-statement models, DCF valuations, LBO analyses, SaaS cohort projections: the AI doesn't just help with the formula. It decides the structure, writes the formulas, applies the formatting, and cites the sources.

What "AI Excel" actually means in 2026

When someone searches "AI Excel" today, they're looking for one of three things:

• Formula assistance: writing, debugging, or explaining Excel formulas. This is table stakes. Every major AI tool does this now.
• Data analysis: querying existing spreadsheets with natural language. Copilot and Gemini are the main players here.
• Model building: creating financial models from scratch. This is where the real value is, and where the tools diverge sharply.

The first two categories are increasingly commoditized. Microsoft bundles Copilot with 365. Google bundles Gemini with Workspace. Formula tools are free or nearly free. The third category, building models, is where specialized tools earn their keep.

The AI Excel landscape: who does what

Microsoft Copilot is the default for most Excel users because it's already there. It's good at data analysis and reconciliation, and it handles circular references natively. But it ignores investment-banking formatting conventions entirely, and it doesn't ask clarifying questions before building. Wall Street Prep gave it a 4.4/10 in their 2026 financial modeling evaluation.

Claude in Excel is the strongest general-purpose LLM for modeling. It asks thoughtful clarifying questions, provides excellent sourcing and comments, and was the only tool to correctly backsolve EBITDA in WSP's test. But it hallucinates historical data at a rate that makes it dangerous for production work without careful auditing.

ChatGPT can produce Excel files you import, but it isn't integrated into Excel. The output is functional but chaotic in formatting. WSP scored it 2.5/10.

Specialized tools like EBITDAI take a different approach: instead of relying on a general-purpose LLM to figure out financial modeling on its own, they wrap the model in a doctrine layer that enforces IB formatting, sign conventions, balance roll-forwards, and domain-specific playbooks. The result is output that looks like a banker built it, not like an AI guessed at it.

What actually matters when choosing an AI Excel tool

After testing every major option, here's what separates useful from dangerous:

• Data accuracy. Can you trust the numbers it pulls? Most tools hallucinate data. The best ones either cite sources or let you upload your own data.
• Formatting discipline. Does it follow IB conventions? Blue inputs, black formulas, bold subtotals, clean spacing. If your MD wouldn't accept it, it's not useful.
• Clarifying questions. Does it ask about forecast preferences, revenue segmentation, and layout before building? Or does it just guess?
• Iteration quality. Can it fix its own mistakes when you point them out? First drafts are never perfect. How well does it respond to feedback?
• Cost structure. Are you paying per token, per seat, or flat? Heavy modeling sessions burn tokens fast. Know what you're paying before you start.

Where AI Excel goes from here

The gap between "AI that helps with Excel" and "AI that builds in Excel" is widening. The general-purpose tools are getting better at analysis but aren't improving at construction. The specialized tools are getting better at construction but still need human review for data accuracy.

The likely end state: AI agents handle the structural work (model architecture, formula writing, formatting, sensitivity tables) while humans handle the judgment work (assumptions, data verification, strategic interpretation). The analyst's role shifts from builder to reviewer, which is exactly where the leverage should be.

For finance teams evaluating AI Excel tools today, the advice is simple: use the general tools for analysis and quick questions. Use a specialized tool for building models. And always, always verify the data before you send anything to your MD.