
If you’re not getting results with AI, it’s probably not your fault.
Many people in the finance industry aren’t seeing the outcomes they expected. That’s what I’m hearing from a CPA Australia project I’m involved in, as well as from conversations with finance leaders.
Why not?
Isn’t AI supposed to revolutionise the way we work?
It turns out, AI is the smartest dumb person I know.
It’s not actually intelligent!
AI predicts what it thinks you want—but most of the time, it doesn’t really know.
It doesn’t stop to ask questions.
It’s armed with vast amounts of information and wants to give it to you, even if it has to make some of it up (thanks to what’s known as an AI “hallucination”).
It’s book-smart but street-stupid.
The best description I’ve heard is that AI is like a book-smart intern without any real-world experience.
Thankfully, like any great intern, it’s willing to learn.
The key to getting results with AI is training it.
I’m working on an AI project with CPA Australia and have been speaking with finance leaders about their challenges. The finance industry is struggling to find practical use cases for AI!
How to train AI
If you ask AI a broad question, you’re unlikely to get a useful answer.
Let’s break this down with an example from my own life: when my wife asks me to make dinner.
If she says, “Alan, can you make dinner tonight?”
I could make anything. She hasn’t specified what she wants, so I might serve up a full Irish breakfast. (For the record, I’m a big fan of breakfast for dinner.)
The next week, she might get more specific: “Alan, can you make steak and potatoes for dinner tonight?”
That’s a bit better, but still vague. She didn’t say what type of steak or how the potatoes should be cooked. For instance, she doesn’t like rump steak, but I love it!
Now, if she gives me an example, I’ll know exactly what she wants: “Alan, for dinner tonight, could you make the steak dinner you did last month with the porterhouse, creamy mashed potatoes, and peppercorn sauce? I’d like a 250g porterhouse cooked medium-rare.”
Very specific.
She could also add a constraint: “No onions this time, please.”
Not a mind reader
Although my wife might wish I were a mind reader (and I wish she were too), if she wants a specific meal, she needs to make a specific request.
AI—whether it’s ChatGPT or any other tool—is just like that. It’s your new intern, and to get useful outputs and answers, you need to be specific.
Training AI is all about refining your input: starting with a broad idea, narrowing it down, giving examples, and setting constraints.
To help you get more out of AI, I’m hosting an online event:
The session will be recorded, so even if you can’t attend live, register and I’ll send you the recording and resources.
Here’s what I’ll cover:
- What AI can (and cannot) do for finance today
- Processes you can easily automate with AI
- The secret to using AI to strengthen stakeholder relationships
- Practical ways AI can help you influence decisions
- FAQs on data security
Here is the registration link.
Thanks,
Alan
Alan “What’s for Dinner?” Cameron-Sweeney
PS: There’s a bonus for those who attend the live session!