Read time: 5 minutes

Welcome back.
Glad you're here. Your competitors are not. That's their problem!

Most companies treat AI failures as overhead to eliminate.
IKEA treated theirs as a product roadmap.
The difference was worth $1.4 billion.

IKEA's chatbot, Billie, handled 3.2 million customer service interactions.
It worked. It resolved 47% without a human.
Billie took on nearly half of all inbound calls, which put 8,500 jobs at immediate risk!
Most companies would have declared that a win, logged the cost savings, and moved on.

IKEA asked a different question:
“What could these people do if we freed them from the work a chatbot can handle?”
IKEA looked at the other 53%."

Almost all of the unresolved tickets were customers asking for help with room layouts.
Not a support failure. A demand signal for a service IKEA didn't offer.

They retrained all 8,500 call centre workers as interior design advisors.
A genuinely different role, requiring new skills, new knowledge, and proper training investment.
The service starts at £25 for a 45-minute video consultation. The interior design channel already existed, but the reskilling scaled it dramatically.
The result: $1.4 billion in revenue.

A cost centre became a revenue generator.

It is worth being accurate here: IKEA's parent, Ingka Group, has also announced around 800 corporate office redundancies this month as part of a broader restructuring. Layoffs are real and the anxiety around AI's impact on jobs is legitimate.

But the reskilling story proves an alternative exists.
The question most organisations are asking is how fast can we automate.
The more valuable question is what your people could do if the repetitive work disappeared.

For finance leaders specifically, that question is not abstract.
It is about your team's time.
What are your analysts doing today that a well-prompted AI could handle in minutes?
And what would you want them doing instead?

A thread from @PEoperator on X made the rounds in March and it is worth unpacking for finance leaders specifically.

A CFO at a PE-backed business walked into a board meeting with no model, no projections. Just a conviction. He had spotted an abandoned e-commerce project, seen a competitor dominating online, and noticed customers were already asking for it. He pushed for a £26k portal launch with nothing in a spreadsheet to back it up.

Within 18 months it was 15 to 20% of total sales.

His point: the spreadsheet phase of your career is when you learn how the sausage is made. It is a foundation, not a destination. And too many people never leave it.

AI is accelerating that shift.
Anything you can model, AI can model faster.
What AI cannot replace is the operator who reads the room and decides.
The person who stops asking "what does the model say?" and starts asking "what do I believe, and what am I going to do about it?"

The problem is most finance leaders are prompting AI the same way they used to use a model: plug in inputs and wait for an output.
That produces average answers, because you are asking for the average of the internet.

The framework I teach in my AI for Finance Leaders course changes that.
It starts with your judgment, not the machine's.

RACEF: Role, Action, Context, Examples, Format.

The E is the part almost everyone skips.
Examples show AI the tone, structure, and standard you expect.
Without it you get generic.
With it, the output sounds like you, matches your team's standard, and saves the rework.

Here is what it looks like in practice.
Instead of "Write a variance commentary for the board pack," try this:

Role: You are a senior FP&A analyst at a mid-market manufacturing business.

Action: Write a three-paragraph variance commentary for the monthly board pack. Address revenue, gross margin, and overheads. Flag the two most significant movements and give a recommended action for each.

Context: Revenue is £2.1M, 4% below budget. Gross margin is 41%, up 2pp versus prior year. Overheads are £800k, 6% above budget due to one-off legal costs.

Examples: [Paste last month's commentary here as a style reference]

Format: Three paragraphs, written for a non-finance board, 250 words maximum.

The before and after difference in output quality is not subtle!

This is from Module 3 of my course.
The full version includes a 64-prompt library, a variance commentary template bank, and RACEF walkthroughs for forecasts, decision memos, and board pack narratives. Currently available at a discounted rate for a limited time. Check it out here.

Three quick ones worth knowing.

Claude now works inside Microsoft Word.
The add-in lets you ask questions about your document, rewrite selected text, and accept or reject AI edits through tracked changes, all inside Word.
It joins the existing Excel and PowerPoint add-ins to complete the Office suite. Currently in beta for Team and Enterprise users, available through the Microsoft Marketplace.

The M365 connector for Claude is now available on every plan, including free. Claude can read your Outlook, search your SharePoint, and pull from Teams chats without you uploading a single file. Microsoft Copilot charges £16 per user per month for essentially the same access. That is a much harder sell now.

Anthropic published its official prompting best practices this week.
Worth bookmarking if you are building internal AI workflows or training your team on how to get consistent output from Claude.