How Do You Stop AI Spend From Spiralling?

Every quarter, CFO Techstack brings together the FLAC, the Finance Leaders Advisory Council. It’s a small group of finance leaders from across the community who compare notes on where the stack is heading. This quarter’s topic was AI.

Last week we shared their take on build vs buy. This week is about what happens once everyone’s already using it: we have a habit that’s getting harder to break, and a bill that keeps getting bigger.

Is AI the new cigarette?

When it comes to the cost of AI, the first thing to look at is how much we’re relying on it.

One finance leader, whose scepticism around AI continues to grow, said:

“I kind of feel like AI is the new cigarettes here. We’re going to get addicted to doing things a certain way (the easy way), and the price is going up.”

For most of us, that’s fair to say. We’re not addicted to the tools: Claude, ChatGPT, Gemini or Copilot. We’re addicted to the quick win.

Ask AI to pull data from three systems and format a report, and you get a result in seconds. That’s genuinely useful. But when we’re using it to write a two-line email or check our grammar, it’s a different story. As another member put it:

“It’s like using a Ferrari to go around the corner to buy a loaf of bread.”

Are we kidding ourselves if we think we can keep driving that Ferrari rent-free forever? The companies behind these tools are raising enormous sums of money, and they’re likely subsidising our usage (for now). At some point, they’ll come knocking.

If we’ve built our workflows around them, or forgotten how to write that email ourselves, we won’t have much choice but to pay.

What does “spiralling” actually look like?

For some members, the bill didn’t wait for the vendors to come knocking. Usage alone was enough.

One member watched their AI spend jump from US$5,000 a month to US$40,000 a month in about four weeks.

It was a shock to the system. But they also saw it as the predictable result of giving people access to these tools, and encouraging them to use them, without a framework or guardrails.

What actually brought the number down?

So what does a framework actually look like? The good news is that the member who’d gone furthest over budget had also found some practical ways to get on top of it.

Three changes in particular made the difference:

Better prompting. Replace vague requests (“write me an email back to this investor”) with specific, bounded ones (“draft an email explaining why this timeline has changed by two weeks, professional, empathetic, under 150 words”). That means fewer tokens burned and less AI slop to clean up afterwards.
No personal use. No more asking the model what the weather’s like in Bali (after all, it's always 29 with a good chance of rain).
Match the model to the task. Not every job needs the most expensive option available.

Together, those three changes took that business from US$40–50K a month back down to around US$12,000.

Changing behaviour works, but it relies on people sticking to it. So several members had also added a simpler guardrail: a card with a hard spend limit that freezes automatically past a threshold. It doesn’t rely on anyone remembering to check.

Is it worth it?

Getting spend under control is one thing. Knowing whether it’s worth it is another, and for most members that question is still open. Most are still in an adoption phase: get usage up first, justify it later.

One engineering-heavy business had set itself an internal bar. If AI costs $40,000 a month, it should be doing the work of roughly four people, not one.

Whether that’s the right yardstick, the jury is still out. As one member put it about ROI more broadly:

“If it’s something new you weren’t doing before because you can now do it with AI, then... you don’t have anything to compare it against.”

Before your AI bill surprises you

Nobody in the room had this fully solved. But while the ROI question stays open, you can still make sure the cost side doesn’t get away from you. If you’re not there yet, this is where we’d start:

Know your number. Who can see total AI spend across every tool and card today?
Set the rules before access widens. Say what it’s for, what it isn’t for, and which model to use for what.
Put a hard limit in place. Use a spend cap that freezes automatically rather than an alert someone has to act on.
Decide what “worth it” means now, even roughly, before you’re asked to justify it.

___

Spend is the visible cost. The less visible one, which never shows up on an invoice, is what happens to institutional knowledge when a business scales AI without anyone owning the overall picture. We'll share how the FLAC are thinking about IP next.

What’s your AI spend doing? Come and share how you’re keeping it in check in The Stack Exchange.

‍

COMMUNITY INSIGHTS

🎤 How I stacked it

Luke Simpson, Finance Director at JENKI
Finalist: Finance Leader of the Year for 2026 🏆

JENKI is an omni-channel matcha brand. They sell matcha drinks, powder, and bakes across bars, online and wholesale. With seven bars in London, they have ambitious plans to scale to ten by the end of the year.

As Finance Director, Luke oversees the financial operations of the business, from reporting and controls to supporting strategic growth.

Visit JENKI

Here are the apps they run:

Xero is the core accounting platform the rest of the stack feeds into, giving a consistent financial foundation across bars, online and wholesale. The team has started on small tasks with the Xero MCP.
Mayday automates prepayments and accruals, bringing more of month-end into one place.
Lightyear automates AP across seven bars and growing, saving the team countless hours.
EasyMonthEnd keeps management accounts on time and stops anything being missed at month-end.
Syft produces management reports at a strong price point, while Excel still handles one-off, detailed analysis.
Ramp runs spend management and employee cards. It was chosen for tighter expense controls and fast syncing of expenses into Xero.
Staffology runs payroll, with a self-coded workaround to split costs by site and revenue centre.
Deputy manages scheduling and rotas across all seven bars, handing off to Staffology for payroll.
Link My Books brings online sales data into Xero, reconciling transactions without the manual work.
Tenzo brings hospitality data together to show how each site is performing. With its MCP, analysis that used to take half a day now takes minutes.
Bombe AI predicts revenue for the bars and online, and helps choose new locations. In hospitality, being on the wrong side of the road can cost thousands.
Claude connects to the MCPs of other apps in the stack, so data flows through correctly and securely.

See JENKI's full stack.

If you want to showcase your stack contact jack.thiel@getmayday.com

‍

GIVE ME HOPE

Reader's question:

“We’ve got a fundraise coming up in six months. As the finance lead, what do I need to be thinking about now?”

Felix's answer:

Six months sounds like plenty of time, but it will go quickly. Once you take away the time spent on 'business as usual' there will only be a few weeks (of total time) where you can get deep on fundraise preparation.

Before jumping into my answer, I will say the fundraising process is a great opportunity to connect with your CEO/Founder. You will be riding right alongside them throughout this process, so make the most of it.

Start with the story

Ensure you are clear on the story; how much you’re raising, the 'why', the 'why now' and what it unlocks. Then make sure the numbers back it up. A gap you find now is a strategy conversation. A gap investors find is a problem.

Know your metrics inside and out

Work out which numbers investors will focus on for your model, such as growth, margin, burn and retention. Calculate them the same way every time, and be ready to explain the odd months.

If the board pack and the pitch deck disagree, someone will notice. So take your time, be diligent... attention to detail matters (maybe more than ever).

Build a model you can defend

Keep it simple, tie it to your actuals, and be ready to explain every assumption. Investors will challenge the assumptions more than the outputs, so be sure to have a justification for every one of them.

Get your house in order

Monthly accounts: on time, reconciled, consistent.
Tax and compliance: nothing overdue or unresolved.
Cap table and contracts: documented and accurate.
Data room: start it now, not when the term sheet arrives.

Watch your runway

It's common for raising capital to take longer than you expect. So if you're 6-months out now, you might be 9-months, 12-months (or further) away from having cash in the bank.

So, before you go too far with anything else, know your cash position. Know what happens if the round takes an extra quarter, and which levers you’d pull to keep things moving forward.

Negotiating with little cash left is puts you in a weak position.

Good luck with the raise!

Have a question you’d like answered? Submit your questions here.

‍

EVENT

"Year End Close: Community Drinks" 🍻🌎

Year end is close, and we're celebrating it in person, around the world!

If you call any of these cities home, register and come along. We'd love to catch up properly.

Cape Town, Thursday 12 November: Book your spot
Adelaide, Thursday 3 December: Book your spot
London, Thursday 3 December: Book your spot
Melbourne, Wednesday 9 December: Book your spot [sponsored]
Sydney, Thursday 10 December: Book your spot [sponsored]
Perth, Thursday 10 December: Book your spot

Note, spots are limited so please only register if you intend to attend.

‍

THE STACK EXCHANGE

🧵 This week's top threads, from The Stack Exchange

The Stack Exchange is the place to connect with peers, stay on top of the latest apps and industry news, and work through the challenges that don't have an easy answer.

Want to join the conversation? Sign up free here.

‍

NEW IN THE WORLD OF CFO SOFTWARE

🗞️ News from the stack-o-sphere

Caveman app launches with promise to help reduce AI token spend. Read more.
Archa, Upcover & Primary partner with Rachel Wong on 'CFO Hackathon' event in Brisbane. Register here.
Xero now connects with Gemini. Bring live financial data into your Gemini conversations. Read more.

‍

WHAT THE DATA SAYS

📊 Stat of the week

32% of CFOs report that their organisations exceeded their AI budgets by at least 10% in the past 12 months.

MORE OF THE GOOD STUFF

And lastly, our top picks!

🎧 Podcast: The AI Exhale. 5-minute AI updates. Listen here.

📝 Article: You need a new CFO | Andreessen Horowitz. Read here.

😆 Joke: The CEO asked for a full ROI analysis on our AI spend. It's all good, I've got Claude on the case.

‍

Why not forward this newsletter to someone you think would enjoy it?

‍