Excel + AI is a Mess (Until You Do This One Thing)

Excel + AI is a Mess (Until You Do This One Thing)

AI is transforming the way we work. It’s writing meeting notes, generating client emails, summarising compliance docs, and helping firms stay productive with half the admin.

But when it comes to Excel-heavy workflows? That’s where things start falling apart.

Accountants are seeing more misses than magic.

Across the board, professionals are testing AI tools like ChatGPT, Claude, or Copilot to help with spreadsheet tasks — financial models, cash flow schedules, multi-tab trackers.

The hope? Upload the file, ask a smart question, and let the AI handle the grunt work.

The reality? Footing errors. Missed links. Broken logic. Summaries that don’t actually reflect what’s happening in the sheet.

As one advisor put it:

“ChatGPT has been a game changer for writing and analysis. But when it comes to spreadsheets, I’m seeing more misses than wins.”

So why does AI struggle with Excel?

Most AI models are built around language. They process context, tone, flow, and intent.

Spreadsheets, on the other hand, are logic-based. Structured data. Relationships between tabs. Custom formulas. Cell references. Conditional formatting. Lookup tables. Pivoted data.

That structure — which makes Excel so powerful — is what makes it hard for general AI to parse.

The AI doesn’t “see” your spreadsheet the way you do.

• You see a three-way forecast with multiple assumptions flowing into a rolling cashflow.

• The AI sees a blob of data, disconnected sheets, and no clue where to start.

Even if you upload the file and ask a seemingly straightforward question — “Summarise the financial model” or “Tell me what’s driving cash burn” — the output often lacks accuracy, nuance, and trustworthiness.

And that’s a big problem when you’re the one responsible for the numbers.

So what does work?

Turns out, the people getting value from AI in Excel are taking a completely different approach.

They’re not expecting AI to magically understand the spreadsheet. They’re structuring the inputs first, and guiding AI step-by-step.

Here are the key strategies that actually work — pulled from a mix of accountants, automation nerds, and AI early adopters.

1. Flatten the file: Don’t expect it to analyse multi-tab logic

Multi-tab files break AI. Especially when there’s no context or naming convention to guide it.

If your file has 6–10 sheets and you want a clean answer, you’re better off combining relevant data into a single dataset first.

Even just linking key numbers into a summary tab with clear labels helps AI work more effectively. The goal is to reduce friction. Give it one structured table, and then ask your question.

Don’t: “Analyse this workbook and tell me if anything’s wrong”

Do: “Here’s a single sheet of reconciled GL entries. Which rows are missing a department tag?”

2. One step at a time: Stop prompting broad questions

AI falls over when you go too big.

“Tell me what’s going on in this spreadsheet” is too vague. Instead, narrow the focus.

Break the task into small, sequential actions:

• Step 1: Identify negative cash movements by week

• Step 2: Categorise by GL account

• Step 3: Group by vendor

• Step 4: Highlight any weeks over a $50k threshold

You get better answers when you work with the AI — not against it. This is known as chaining prompts, and it works best in spreadsheet analysis.

3. Use AI for assistance, not interpretation

Expecting AI to be the analyst is where most people go wrong.

Instead, use it for formula writing, VBA snippets, column cleaning, or text standardisation.

Here’s where AI can genuinely help:

• Write a complex IF/AND/OR formula in one go

• Suggest a SUMIFS or INDEX MATCH approach

• Flag data inconsistencies (e.g. different formats in a column)

• Convert messy text into proper case or date format

• Translate spreadsheet jargon into plain English

Think of AI as your spreadsheet intern — helpful, fast, but needs very specific instructions.

4. Validate with QA tabs inside the file

This one’s for those who really need to trust the output.

Some advanced users are creating QA tabs in the file. These are logic check sheets that mirror what the AI is doing, then verify the results.

For example:

• You ask AI to apply a pricing logic to all rows

• The script outputs a new tab with those prices

• Then it cross-checks against expected outcomes in a QA tab

You can also prompt AI to explain each transformation — “Show your working” — so you can audit it before sharing with clients or using in reports.

5. Use Google Apps Script or VBA to control the flow

This is where things move beyond basic prompts.

If you're confident in scripting, you can connect Excel or Sheets to OpenAI’s API directly. This means:

• Set a defined range

• Pre-structure the data

• Trigger the AI model to perform a very specific task

• Output the result into a new tab or range

All without third-party tools. All under your control.

This approach gives you transparency, repeatability, and better handling of sensitive data.

Bonus: you can store your prompts inside the script so every time the model runs, it’s using your logic, not guessing from scratch.

What most accountants are missing

The reason so many accountants and bookkeepers are struggling with AI in Excel is simple:

You’re expecting too much, too soon.

You’re treating AI like a plug-and-play analyst, when really, it’s a flexible assistant — if you give it the right structure.

There’s a gap in the market right now for real-world, accountant-level tutorials on AI + spreadsheets. Most of what’s out there is surface-level or marketing fluff.

One exception? AutomationTown.io — a community built around real automation workflows for accounting teams. It’s not hype. It’s how-to.

Follow Chad Davis on LinkedIn for more tips and tricks when it comes to AI in accounting.

Where to from here?

If you're using AI in spreadsheets, here's your new checklist:

• ✅ Flatten the data into one tab

• ✅ Be specific with your questions

• ✅ Break down analysis into bite-sized steps

• ✅ Use AI for assistance, not interpretation

• ✅ Add QA where accuracy matters

• ✅ Script it where repeatability matters

This is the one thing most people skip: Structuring your input before expecting a smart output.

Final thoughts

AI is already reshaping the way accounting work gets done.

But Excel isn’t just another app. It’s the backbone of finance, reporting, and decision-making. You don’t want to play fast and loose with your spreadsheets.

So don’t rely on generic AI advice. Build the right scaffolding. Understand what the AI can’t do. And then give it just enough structure so it can help.

If you’ve got a workflow, prompt, or script that works — share it with us. We’ll feature it in an upcoming post from The Firm and help others get smarter, faster, and more accurate with the tools that are already here.