I want to tell you about something that happened on a call I had recently with a finance leader. She was frustrated because every time she tried to use AI for variance analysis, the output was, in her words, "generic rubbish."

So I asked her to show me what she was typing into ChatGPT.

It was one sentence. One.

She hadn't uploaded any data, hadn't given any context about her business, and hadn't shown any examples of what good analysis looks like in her company. And she was blaming the AI for giving her a bad answer.

I see this all the time in the AI Finance Club. Finance pros spending 20 minutes trying to create the perfect prompt, deciding whether to say "analyze" or "evaluate" when the problem is not giving AI anything to work with.

So when I sat down with Shawn Kanungo (who advises enterprises globally on AI and has spent 12 years at Deloitte in management consulting), he said something that made a lot of sense.

"I never come up with a prompt myself. I always ask AI for the prompt and it will come up with a better prompt than you."

Then he broke down why most people are getting bad results.

And guess what… It has nothing to do with prompt engineering.

​

​
Your Prompt is not the Problem

I’m sure you’ve been in this situation where you stare at an empty AI chat and are thinking super hard about what will give you the best results.

Maybe you Google "best prompts for financial analysis." Copy/paste, get an output that is not good, and then decide AI is not ready for ‘serious’ finance work.

I know this because I've watched hundreds of finance professionals do exactly this in my training sessions. They treat AI like a search engine and then wonder why it gives them the most expected results.

But Shawn put it really simply.

"Using these large language models comes down to four different things. Number one, it's the context. The more context that you give it, the better. So I try to give it as much documentation and as much verbal diarrhea as possible."

Your prompt is not the problem. It’s that you're giving it a one-line request with zero background and expecting it to deliver what a senior analyst would.

This is like handing a new hire a blank Excel file and saying "do the analysis" without any brief, any sample output, or any explanation of your business.

​

​
The 4-Input Method

Shawn's framework reverse engineers how you should be thinking about AI.

Most people start with the prompt (the thing they stress about most). Shawn starts with context (the thing you already have) and leaves the prompt for last.

Here are the four inputs, in order of importance.

Input 1: Context

This is the big one. Upload or paste everything that's relevant to what you're working on. The Excel file, the prior month's report, the email thread from your business partner, a screenshot of your dashboard. Don't try to make it perfect, or super clean. Just paste it all in there.

Think of this like briefing a contractor you've just hired. You wouldn't spend 30 minutes writing the perfect email to them. You'd send them the task, the budget, examples of what "good" looks like, and a clear description of the problem.

Input 2: The Question

Spend 30 seconds writing down the business problem you're trying to solve. Something like:

or,

Input 3: The Model

Use whichever LLM you have access to. Just make sure it's a secure tool in Extended (Claude), Thinking (ChatGPT & Copilot) or Pro (Gemini) mode. The important thing here is that you're using a reasoning or 'thinking' model. You can read more about AI in thinking mode here.​

And here's something I always tell people: don't overthink this.

The tool matters way less than the context you give it. Using one tool with great context will beat another tool without context every single time.

Input 4: The Prompt (Let AI Write It)

This is the part that surprises everyone. Once you've loaded your context and clarified your question, you type something like:

AI will write a better, more structured prompt than you would have come up with yourself.

Review it, change it if something doesn’t look right, then use it.

That's it.

​

​
How to Use This on Monday Morning

So here's what I want you to do with your very next AI task.

Step 1: Before you type anything into the chat window, gather your materials. The spreadsheet, the prior report, the email from your business partner with context. Upload all of it.

Step 2: Write down your business question in one or two sentences. What problem are you trying to figure out?

Step 3: Open your preferred tool (ChatGPT, Copilot, Gemini, Claude) and dump everything in.

Step 4: Type "Based on everything I've given you, write me the best prompt to [your question from Step 2]."

Step 5: Review the prompt AI gives you, change it if you need to, then use it...

Something else cool:

Shawn mentioned he doesn't even type his context anymore. He uses a tool called Whisper Flow (which is a voice-to-text app for your phone and laptop) to just talk through what he needs.

You can do the same with your phone's built-in dictation or any speech-to-text tool. Talking naturally gives AI richer context than typing because you explain things the way you'd explain them to a colleague, and that's exactly what AI needs.

The One Thing to Remember

So if you take one thing from this newsletter, it is this.

The people getting incredible results from AI are not better at writing prompts than you. They just feed AI more context, get clear on the question, and then work with AI until they get they outputs they want.

Remember, your prompt is not the problem.

Your process is where the power is.

​

Best,

Your AI Finance Expert,

Nicolas

​

P.S. What did you think of this? Hit reply and let me know if you try the 4-Input Method this week (I read all replies).

P.P.S. This newsletter is based on my conversation with Shawn Kanungo where we went much deeper into Claude Skills, synthetic data, AI audit coaches, and more. Watch the full video here The Blueprint to Using AI for Finance in 2026 (ft. Shawn Kanungo).

​

​