49 out of 50 CFOs missed this
Two weeks ago, I ran a live exercise with 50 finance pros.
I gave them all the same supermarket dataset and the same prompt template. Then I asked them to share their screens.
Asif went first. He asked his AI to build a promo ROI scatter plot. Where the bubble size related to the promo dollars spent.
Salmon sat in the top-right corner. The single most profitable product the supermarket was promoting.
I could not believe this. I asked him to scroll back up so I could see better.
|
|
Nobody else on the call had run that analysis.
I said: "Who could have thought of that, from the 50 of us, even with all of our experience combined?"
This is the part that is super important, that’s a lot more than just ‘faster with AI’.
The angle in your data that only YOU can find.
So today I am going to give you a framework to do just this.
70% of AI Success Depends on the Organization, Workforce, and Skills
BCG (Boston Consulting Group) published research this month that says that only about 10% of AI's value in finance comes from the model itself. 20% from the platform. The other 70% is the team's skill at using it.
This is a big problem for CFOs. Prompting can only take you so far.
I see it every week. Someone uploads a file. They type "analyze this." Then AI gives them back a summary. The things you can already see from 5 minutes looking at an Excel file.
Like I said in the masterclass: "If somebody on your team did this work – just basically repeating in an email what was in the Excel file – would you be happy?"
The reason AI stops there is not the model. It's that humans stop there. AI follows the analysis patterns we've trained it on, and most historical financial analysis is descriptive (what happened) and diagnostic (why it happened).
So AI defaults to the analyses your team was already doing manually. The decisions don't get better. They just get faster.
Doing things faster just creates more work.
Making better decisions takes you to a new level.
Gartner's 16-year-old analytics ladder
There's a framework that's been the standard in business intelligence since the early 2010s. Gartner's ‘Analytics Ascendancy Model’.
|
|
There are four rungs on the ladder:
- Descriptive (what happened)
- Diagnostic (why)
- Predictive (what will happen)
- Prescriptive (what should we do
Most humans look back (what happened and why). The top two (what will happen and what should we do) are where your strategic value sits.
And they're the ones your team is least practiced at requesting from AI.
Salmon-as-top-promo was prescriptive: it told the CFO on my course where to put next quarter's promo dollars.
Just like the BCG's report says:
You're not the analyst anymore. You're the one deciding which analyses are worth running (and finding new analysis points that nobody else thought of before).
The Ladder Analysis Method
By the end of this section, you'll have a prompt structure you can paste into ChatGPT, Claude, or Copilot today. The same file, but four times the analytical depth.
|
|
1. Brief AI like a Data Scientist
Start with role and goal in one line.
|
|
Not "analyze this file."
The role and the decision come first.
2. Name all four rungs explicitly:
Add this line to every analysis prompt:
|
|
Naming them is what stops AI defaulting to the bottom two.
3. Ask for two analyses your team wouldn't think of:
Add:
|
|
This is the type of line that produced the salmon scatter plot. It forces AI off the well-trodden boring path it’s been trained to default to.
4. Make AI pick the visualization
Add:
|
|
Asif's scatter plot worked because the AI chose bubble size for promo dollars – a choice no human in the room made.
5. Review like you'd review a junior analyst:
AI gets you to 80%. You then bring the business judgment.
So pick the two analyses that support (or change) a decision.
Then get rid of the rest. The only job here is to find the analysis worth acting on.
|
|
|
|
|
|
The One Thing To Remember
For the last 20+ years, the best finance professionals were the ones who could find an insight in a spreadsheet faster than anyone else.
For the next 20+, the best ones will be the ones who can get AI to surface the insight nobody on their team would have thought to look for otherwise.
Same data. Different prompt.
Salmon was sitting in that supermarket dataset the whole time. It just took one CFO, using the Ladder Analysis method to find it.
The question for your team this quarter isn't "how do we use AI to work faster?"
It's "what’s in our data that we have never thought to ask about?"
Best,
Your AI Finance Expert,
– Nicolas
P.S – What type of analysis are you struggling to use AI for? Hit reply and I’ll see if I can help (I read all replies)
P.P.S – We had 4,595 register for our last AI CFO webinar where I presented live from Namibia where I am on holidays with my family. I take time out of my holidays, as you take time out of your schedule to join me in learning AI for Finance. I'd love to have you join me for my next masterclass here.
|
|