So, you have spent hours typing into ChatGPT, Gemini, Copilot or Claude.
You’ve had a really long discussion, and at the end you finally get the output you wanted from the beginning.
If you're still doing this every time you use AI, you’re going to go crazy!
|
|
You're not doing anything wrong. You're just stuck in what I call a ‘Prompting Loop’, where every AI conversation feels like a first meeting, when it should feel like talking with a colleague you’ve worked with for a long time.
So, today I want to show you how to fix this.
And yes, becoming a better at prompting helps, but what’s easier is building reusable AI assistants that remember your context, your company, and your quality standards, so you get the right output on the first message (not the 87th).
Plus, it can be as simple as asking 1 simple question.
The Prompting Loop
You started using AI to save time. But every time you sit down to use it, you're spending the first 20 to 30 minutes just getting AI back up to speed on who you are and what you need.
Think about it this way. You would never manually format a monthly P&L from scratch 12 times a year. You build the template once, then fill in new data each month.
But with AI, most people are rebuilding the template from nothing every single time they open a new conversation.
|
|
This is the biggest difference between people who say "AI saves me hours" and people who say "I tried AI but it's too much effort."
Both groups have the same skills. But, one group is still prompting from scratch, the other is building assistants…
Build Once, Use Forever
So here’s a simple fix that you can try right now.
When you finally get that perfect output after an hour of back-and-forth,
Capture that conversation by asking one final question.
AI will write you a detailed prompt that captures everything it learned during your conversation.
You save that prompt. And then you turn it into a reusable AI assistant.
(full step by step in the next section)
Now, the term for these assistants is different on every platform, and this confuses a lot of people (plus it’s changing all the time).
|
|
Whether it's a GPT, Gem, Agent or Project – They all do the same thing, just with different branding.
You write the instructions once, and every time you open that assistant, it already knows your context.
You don't need to re-explain anything or waste messages getting AI up to speed. You go from spending 45 minutes teaching AI to spending 2 minutes getting exactly what you need.
Here's how to build them…
How to build a custom AI assistant
Here’s my 6-step process. You can do this in ~20 minutes, and you’ll use the result for months (or years).
|
|
Step 1: Pick one mission.
Don’t build a “Finance Helper” that does everything.
Build a “Variance Commentary Writer,” an “Expense Categorizer,” or a “Board Deck Summarizer.”
One assistant, one mission. When it does that brilliantly, you can build the next one.
Step 2: Generate your system prompt.
Open a new conversation chat or scan all your previous chat conversations where the output is exactly what you want. If you like the output that was generated, end it by asking this.
Save that output somewhere, you are going to need it in a minute.
Step 3: Paste those instructions into the builder.
Open your platform of choice.
- In ChatGPT, go to "Explore GPTs" and click "Create."
- In Gemini, go to "Gems" and click "New Gem."
- In Copilot, open the Agent Builder or Copilot Studio.
- In Claude, create a new Project.
Give your assistant a name.
Then, you're going to use the prompt from the previous step now as your instructions. Paste it into the builder.
You can always improve or further customize the instructions later.
Step 4: Add knowledge files.
One more thing that makes a big difference is to include knowledge files.
Upload 2–3 examples of outputs you loved, and 1–2 examples you don’t want.
Plus, add any additional context documents that will help the AI understand more about you and your task.
This is like onboarding a new team member. Showing them what "good" looks like is always faster than describing it abstractly.
Step 5: Test, iterate, and fine tune.
Before using your AI assistant, it's a good idea to test.
Run 10 real test prompts. Aim for 90–95% of outputs matching your standard. If the output is not right, tweak the instructions rather than starting over.
Small changes go a long way. So tweak until you get to the results that you want.
Step 6: Share it (so the whole team works the same way).
Once it's working, share it with your team.
This is where it gets really powerful. Instead of five people on your team all prompting differently and getting inconsistent results, everyone uses the same assistant with the same standards.
What you can do right now
Here is your challenge for this week. Build one custom assistant for one repetitive finance task.
Pick the task where you find yourself re-explaining the same context to AI over and over again. It doesn’t even need to be a complicated task.
Build the assistant. Test it. And the next time that task comes up, see what happens when AI already knows who you are, what you need, and how you like it delivered.
Because the finance professionals who are getting the most from AI are the ones having the shortest conversations with it.
You don't need to be AI's teacher every time you start a task.
Now you can be its manager getting it to deliver results in a fraction of the time.
Best,
Your AI Finance Expert,
Nicolas
P.S. Have you already built a custom assistant? Or is this new to you? Hit reply and let me know (I read all replies).
P.P.S. If you want to see the full breakdown of AI tips for finance, check out my latest YouTube video “100 SECRET tips on AI for FINANCE/ AI for CFO” where I walk through 100 tips you can use right now.