CFO Daily · Wednesday, September 23, 2026

Hi, this is Koko from Koko Knows.

Let's get right to what's new for finance leaders today, because there's a signal out of Toronto that should sound very familiar if you've been running AI pilots this year. At a CFO Alliance roundtable, midmarket Canadian CFOs said the quiet thing out loud: their AI experiments aren't stalling because the models are weak. They're stalling because the plumbing underneath them is broken. Unreliable data-sharing systems, tech spend that was never properly allocated, and, maybe most alarming, shadow AI showing up on personal accounts that nobody in finance can see or track. That last one should get your attention immediately. If people on your team are running finance workflows through consumer AI accounts, you don't have a pilot problem, you have an exposure problem, and you likely don't even know the size of it yet.

Here's why this matters beyond one roundtable. This same pattern is showing up at scale in the broader AI spending story this week. McKinsey, via CIO Dive, found that most enterprises are overrunning their AI budgets on coding agents specifically, and doing it despite weak evidence that those agents are actually delivering ROI. So the pattern is: spend is accelerating faster than measurement discipline, and that gap is happening in engineering budgets today and will happen in your finance and FP&A tooling budgets tomorrow if you let it. Put those two stories together and you get a clear mandate. Before your next budget cycle, you need two things in hand: a documented inventory of shadow AI use across your organization, and a hard ROI gate on any coding-agent or finance-agent spend before it gets renewed past its first cycle. Not a vendor pitch deck. Actual measured output.

Now, a couple of things worth watching on the infrastructure and vendor side, because they change how you should be negotiating contracts and modeling counterparty risk. First, professional-services pricing is shifting under your feet. The commentary out of Certinia and Dreamforce this week points to firms moving away from hourly billing toward fixed-bid, outcome-based pricing, because AI is compressing the labor that used to justify the hourly model. If you're negotiating a consulting or systems-integration contract right now, this is your moment to push for outcome-based terms while you still have leverage, before vendors reset pricing power around "determinism" claims that are genuinely hard for you to audit later. Get the outcome definitions and the audit rights written into the contract now.

Second, if you have any exposure to AI infrastructure players, either as a customer, a lender, or through supply chain dependency, take a hard look at how these deals are actually financed. IREN's latest earnings call disclosed nineteen billion dollars in funding, but sixteen billion of that is prepayments and GPU financing, and it's sitting against just four billion in contracted annual recurring revenue. That's a financing structure that looks impressive on a headline but is thin underneath if demand doesn't materialize as promised. Nvidia just added one and a half billion dollars to its stake in SB Energy, bringing its total commitment there to three billion, alongside a hundred and five billion dollars in residual-value guarantees tied to four and a quarter gigawatts of capacity. Alibaba's expanding its AI chip and data-center build too. Capex across this whole ecosystem is running well ahead of demonstrated enterprise returns. If your treasury or credit teams are modeling exposure to any AI infrastructure counterparty, stress-test the prepayment-heavy structures, not just the headline revenue numbers.

The strategic thread tying all of this together is really about where the moat actually sits. When frontier models are landing with near-identical capability and forty to fifty percent price cuts within ninety minutes of each other, model choice is becoming a commodity procurement decision. That's actually good news for your cost line, but only if you act on it. If your AI budget was built on last generation's token pricing, it's stale and overstated right now. Push vendor renegotiation today. But don't stop there, because the real risk isn't the sticker price, it's untraceable spend and shadow usage that never shows up in a chargeback. Capture the savings, but tie it to better usage tracking and governance controls, or you'll just be trading one blind spot for another.

So here's the through-line for today. The gap between how fast you're spending on AI and how carefully you're governing it is the same gap. Close it with an inventory, a gate, and a renegotiation, not with another pilot.

That's your briefing. This has been Koko from Koko Knows, back tomorrow with what's next.