A hundred years ago, retail pioneer John Wanamaker griped that half the money he spent on advertising was wasted — he just couldn’t tell which half. Most enterprises are running their AI programs the same way today. Except the bill is bigger, it compounds every month, and your board is asking about it.

Executive summary

AI spending is growing faster than any technology budget in corporate history. Gartner projects generative AI software spending to jump 76% year over year, reaching $644 billion in 2025. IDC expects enterprise investment in AI infrastructure to reach $571 billion in 2026. And yet, in Apptio’s research, 55% of business leaders admit they don’t have the information they need to evaluate their technology spend. IDC finds 38% of organizations missing their AI ROI targets — not because AI doesn’t work, but because hidden costs quietly eat the return. The problem isn’t that AI is too expensive. The problem is that almost nobody can connect what they spend to what they get. FinOps for AI is the discipline that fixes this, and IBM Apptio’s platform — Cloudability, Kubecost, Apptio, and Targetprocess — is how you put it into practice. It comes down to three moves any executive can understand: see every AI dollar, give every dollar an owner, and stop bad spend before it happens.

Why your financial controls can’t see AI

Your finance team built its controls for a world where technology costs were predictable. You bought licenses. You signed contracts. You budgeted annually and reviewed quarterly.

AI ignores all of that.

AI costs are consumption-based — priced per token, per GPU-hour, per API call. One enthusiastic team can burn a quarter’s experimentation budget over a long weekend, and you’ll find out when the invoice lands thirty days later.

The spend is scattered, too. Some sits with cloud providers. Some goes straight to model vendors like OpenAI and Anthropic. Some hides in GPU clusters running on Kubernetes. Some is buried inside the AI features of SaaS products you already pay for. Add shadow AI — tools that teams adopt on a corporate card — and no single system in your company sees the whole picture.

So the CFO gets a number. What the CFO can’t get is the sentence after the number: which team, which model, which product, and was it worth it.

Move one: see every dollar

You can’t manage what you can’t see. And with AI, most companies currently see very little.

IBM Cloudability fixes the visibility problem by pulling every AI-related cost into one place — multi-cloud bills, Kubernetes clusters, SaaS subscriptions, and AI workloads together. Usage telemetry — tokens consumed, GPU hours, API calls — flows in through the FOCUS open billing standard and custom connectors. That includes third-party providers: your OpenAI spend shows up next to your cloud spend, in the same view, in near real time.

For companies running their own GPU infrastructure, IBM Kubecost goes a level deeper, with container-level cost detail on shared clusters. GPUs are the most expensive real estate you own. You should know exactly who’s occupying them.

Move two: give every dollar an owner

Visibility without ownership is just a prettier invoice.

The second move is allocation. Every AI cost gets tagged and mapped to a team, a product, a business unit — then it shows up on that unit’s numbers. Showback first. Chargeback when you’re ready. Behavior changes remarkably fast when AI spend stops being “IT’s problem” and starts appearing on a leader’s own P&L.

Allocation also unlocks the question your CFO actually cares about: unit economics. Not “what did we spend on AI?” but “what did each finished piece of work cost us?” Cost per feature shipped. Cost per claim processed. Cost per customer conversation resolved. Those are numbers a business can steer by. Total spend is not.

Move three: stop bad spend before it happens

Monthly cost reviews are how you find out about a problem. They are not how you prevent one. With AI, thirty days is an eternity.

Cloudability Governance moves the control point forward, into the engineering workflow itself. Through its Terraform integration, engineers see the cost impact of infrastructure before they deploy it, and tagging rules, budget limits, and compliance policies are enforced automatically as they build. Finance stops being the department that says no after the fact, and becomes a guardrail that’s simply built in.

For everything that slips through, machine-learning anomaly detection watches spend in real time and flags spikes as they happen — not at month end, when the damage is already three weeks old.

The part most companies skip: proving value

Cost control is only half the job. The other half is knowing whether the spend was worth it.

Apptio’s AI TCO & Usage capability builds the number most companies have never actually calculated: the full cost of an AI initiative — cloud, model vendors, infrastructure, and the people working on it, all together. IBM Targetprocess then connects those initiatives to business outcomes, so you can compare investment plans side by side and defund the ones that don’t return.

Most organizations today review AI projects on enthusiasm. You’ll be reviewing them on numbers. That’s a durable advantage.

What to ask on Monday morning

Three questions will tell you where you stand:

• What was our total AI spend last quarter — across every provider? If the answer takes more than a day to assemble, that delay is the finding.
• Which three teams or models drive most of the spend — and is it trending up or down?
• What does one AI-powered outcome — one resolved ticket, one processed claim — actually cost us?

If nobody can answer, you don’t have an AI cost problem yet. You have something worse: an AI cost mystery. The good news? It’s fixable, the platform exists, and most of your competitors are exactly as blind as you are.

The bottom line

Wanamaker never solved his half-the-money problem, because the measurement didn’t exist. Yours does.

The winners of the AI era won’t be the companies that spent the most. They’ll be the ones who could prove what every dollar bought.

Which one will you be?