Two Ways to Put AI Inside Your Excel Workflow
There are two setups, and most finance professionals use both. Knowing which to reach for is half the skill.
- Copilot in Excel (in-cell). Microsoft 365 Copilot sits inside the workbook and acts on your data directly — writing a formula into the cell, generating a chart, or summarizing a range. Best for speed on live sheets, and it keeps your data inside your organization's governance boundary.
- A standalone model (ChatGPT, Claude, Gemini). You paste in the relevant range or describe the problem, and the model reasons through it. Best for the thinking-heavy work: building a model from a brief, explaining an inherited spreadsheet, or stress-testing assumptions. Stronger reasoning, but you are responsible for what data leaves the sheet.
Do not paste confidential, client, or material non-public information into a consumer AI tool. Use your firm's approved enterprise AI (Microsoft 365 Copilot and enterprise ChatGPT/Claude carry data-handling commitments), anonymize tickers and figures first, or work on the formula logic with dummy numbers and apply it to the real sheet yourself.
AI is a very fast junior analyst, not the analyst of record. It produces the first pass; you own the assumptions, check the formulas against a known case, and take responsibility for the output. Every workflow below is written with that in mind.
6 AI-in-Excel Workflows Finance Pros Actually Run
Write the formula you can picture but can't quite type
Nested INDEX/MATCH, SUMIFS across sheets, dynamic arrays, XLOOKUP with error handling — describe the outcome in plain English with your real column layout and let AI produce the formula, plus a plain-English explanation of how it works so you can maintain it later.
Write an Excel formula for this. My layout: [describe columns, e.g. "A = date, B = ticker, C = sector, D = P&L"]. I want: [plain-English result]. Return the formula using my exact column references, a one-line explanation of how it works, and one edge case I should test it against.
Turn an ugly export into an analysis-ready table
Bank statements, broker exports, and vendor CSVs arrive messy — merged cells, inconsistent dates, text-formatted numbers, stray footers. AI can give you the exact steps (or a Power Query / formula recipe) to normalize it, repeatably.
Here is the shape of a messy Excel export: [describe the problems + paste 3-4 anonymized sample rows]. Give me a step-by-step cleanup recipe (Power Query steps or helper-column formulas) to get it into a tidy table with one row per record, consistent date and number formats, and no stray text. Make it repeatable for next month's file.
Scaffold a model from a brief in minutes
A three-statement build, a unit-economics model, a simple DCF — AI can lay out the structure, the driver rows, and the formula logic from a clear brief, so you start from a skeleton instead of a blank grid. You still own every assumption.
Scaffold an Excel model for [what: e.g. "a 5-year DCF for a SaaS business"]. Drivers I want to control: [list]. Lay out the tabs and rows, specify the formula for each calculated line (referencing the driver cells), and include a sensitivity table on [the 2 variables that matter most]. List the assumptions I must set myself and flag the ones most likely to swing the output.
Find where your model breaks
The value of a model is in its assumptions, and we are all blind to our own. AI makes a tireless reviewer — point it at your logic and ask where it falls apart.
Here is the logic and key assumptions of my model: [describe drivers + paste the assumption block]. Act as a skeptical reviewer. Which 3 assumptions is the output most sensitive to? What is the most optimistic assumption I have made? What scenario would break the thesis, and what would I need to believe for the base case to hold?
Understand a spreadsheet someone else built
Inheriting a 12-tab workbook with no documentation is a rite of passage. AI can map the structure, trace a number to its source, and flag the fragile bits (hardcoded values inside formulas, broken references) before they burn you.
I inherited this Excel workbook. Here is the tab list and the formulas from the key output cells: [paste]. Explain what this model is doing in plain English, trace how [the headline output] is calculated, and flag anything risky: hardcoded numbers inside formulas, likely-broken references, or circular logic I should double-check.
Turn a range of numbers into a paragraph a client understands
Half of finance is translating a table into a sentence a non-analyst can act on. AI drafts that translation from your numbers, in your tone, ready for you to check and send.
Here is a summary table from my Excel analysis: [paste the summary rows]. Write a [length] plain-English commentary for [audience, e.g. "a non-technical client"]. Lead with the one thing that matters most, explain the key driver, and avoid jargon. Neutral, factual tone. Do not add any numbers that are not in the table.
The Mistakes That Get Analysts Burned
AI in Excel is a genuine edge, and it is also where careless use shows up fastest. The failure modes are consistent:
- Trusting a formula you did not test. AI-written formulas are usually right and occasionally confidently wrong. Always test on a row where you know the answer.
- Pasting data you were not allowed to. The convenience is not worth a data-handling breach. When in doubt, anonymize or use the approved enterprise tool.
- Presenting unchecked output as your analysis. The moment someone asks a follow-up you cannot answer, the shortcut is exposed. Know your own model.
- Letting the model invent numbers. In commentary and summaries, explicitly instruct it to use only the figures you provide — and verify.
Every workflow here is accurate as of July 2026 and tested on real work. Six months from now, Copilot will do more in-cell, models will handle bigger sheets, and new patterns will emerge. A static article cannot keep up; a weekly brief can.
AI Finance Brief ships one practitioner-tested AI workflow each week — the exact prompt, the data source, and the output to expect — built specifically for finance desks. It is the lowest-friction way to keep your Excel-plus-AI stack sharp as the tools evolve.
Frequently Asked Questions
One tested AI workflow
for your desk, every week
AI Finance Brief is a free weekly newsletter for finance professionals who want to actually use AI — exact prompts, real data sources, tested on live markets. Start with the free brief; upgrade to Pro when you want the full code and templates.
Free forever. One email per week. Unsubscribe anytime. Read a full sample issue →