Follow the Money: How CIOs Should Talk to the CFO About AI

What is the fastest way to lose AI funding? Walking into a CFO's office prepared to answer only one of their two critical questions: "What does it cost?" and "What do we get from it?"

Technologists typically focus on costs, but CFOs need both answers, and each requires a distinct vocabulary. Swapping the vocabulary in your answers to either question is a huge red flag to a CFO. In fact, doing so is how good investments get killed in quarterly reviews. Industry analysts are projecting that a meaningful share of enterprise AI efforts will miss the mark, and from what I see across the CIOs I advise, most of those misses will look like budget decisions, even though they're really vocabulary problems.

How to talk about value

First, follow the money

The principle I keep coming back to on both questions is the same one I've used for years: follow the money. And always remember that language will change depending on which side of the money you’re describing.

When we were deciding where to apply AI first, we developed a three-question test: What are the company's highest-cost areas where AI can deliver measurable impact? How do we reduce or optimize those costs? How can AI shorten our time to market or generate revenue? Asked in that order, those questions set prioritization and help keep discussions constructive while avoiding political debates between departments.

Second, pick the right bucket

Once you've chosen a use case, the harder problem is describing its value in language that Finance will recognize. Most teams say ‘savings’ when they mean ‘business value,’ and the two words trigger very different expectations once Finance is in the room. AI shows up as productivity, velocity, and savings, and each lives in a different bucket.

I place business value in three distinct buckets. A CIO should be able to tell a CFO which bucket(s) a given AI investment falls into before the meeting begins.

Bucket 1: The velocity bucket. Revenue, time-to-market, product capability. An AI investment here might make IT more expensive, and that's fine if the top line is moving in response. Defending this kind of AI investment is a revenue conversation. Be careful not to dress this up as savings, or it will get defunded.

Bucket 2: The productivity bucket (cost avoidance). The unit rate comes down. The same team does more work for the same cost basis, so the business grows without the P&L growing alongside it. Warning: this is the bucket teams most often mislabel as savings, which sets up the wrong argument with Finance.

Bucket 3: The savings bucket. The P&L line genuinely contracts when headcount, vendor spend, or run-rate costs drop.

Most companies pass through buckets one and two before reaching bucket three, and that maturity curve is fine as long as you’re honest about where you are on it.

All three of these buckets have a valuable business impact. The problem occurs when IT fails to communicate in the language of Finance. Misusing the vocabulary of each bucket is what breaks the conversation. When teams say savings, CFOs hear bucket three. When CFOs hear “bucket three” but see “bucket one math,” your credibility evaporates, and your program loses funding.

The discipline: pick the bucket before you walk into the meeting, defend the bucket you picked with the correct vocabulary, and don't try to defend all three buckets at once.  

A working example

According to the 2024–2025 Intel IT Annual Performance Report, applying this thinking meant focusing initial AI investments on a tight set of four key areas: manufacturing, pre-silicon, post-silicon, and software development, with a centralized, governed platform behind them. By mid-2025, that platform had saved roughly a million work hours, driven by 30,000 monthly users creating more than 7,000 personal assistants across 400+ GenAI use cases, and generated $376 million in business value. Finance reviews validated those numbers by measuring them against productivity and velocity, where the use cases lived. Claiming them as headcount reductions would have been a different argument, and it would have been much shorter. That distinction is what kept the program funded.

How to talk about cost

For a complete assessment, we need to discuss the total cost of ownership (TCO) as well as data quality and governance (which we’ll cover in another opinion piece). Cost is where I see CIOs caught off guard. Deloitte’s recent CFO Insights for AI puts the problem plainly: “As investment in AI rises, finance leaders come face-to-face with the difficulty of calculating a return. Some of AI’s benefits can be hard to measure, while the technology evolves at a speed that can outstrip metrics.” AI pricing doesn’t behave the way enterprise software pricing has behaved for the past twenty years, and the pricing models aren't what most finance teams expect.

Know the four pricing models

• Per seat: Familiar territory for Finance. Easy to forecast, but the costliest when adoption is wide and usage is light, which is what most enterprises end up with.

• Per token or per query: Consumption-based. Scales smoothly until a power user discovers what the tool can do, or an agent runs in a loop overnight.

• Per outcome: A newer model type. Often opaque and easy to underestimate at the time of signing, since it’s tied to vendor-defined success metrics.  

• Hybrid: Where most enterprise deals land. A base platform fee plus consumption. While your base is predictable, your consumption isn't.

Decide who pays

After pricing, the next question is who pays the bill. Whether it’s the user, the team, central IT, or a shared services pool, each choice has consequences for the other side. When the user pays, adoption stalls, and the value never materializes. If central IT pays, the budget line grows faster than IT can defend it at budget time. When the team pays, governance suffers because teams will rationally optimize for their own use.  

Sure, there's no universal right answer. But there is a wrong default I often see: letting the invoice land wherever it happens to end up, and accepting that as the cost owner.

Set controls before the bill arrives

The discipline I’d push on every IT organization right now is to handle AI consumption the way we learned to handle cloud consumption a decade ago. Set budgets, alerts, kill switches, and forecasts before the bill arrives. By the time you're explaining a surprise to Finance, you've already depleted the trust you'll need for the next AI initiative.

What goes wrong most often

The most common mistake that wrecks AI finance conversations is mixing your buckets: making a value argument that's secretly above-the-line, presented as savings, attached to a cost forecast that treats consumption pricing as if it were a flat subscription.  

A clearer, more defensible version sounds like this: This AI investment is in the velocity bucket. It supports revenue, not savings. The cost model is consumption-based, with a quarterly cap and a monthly review with Finance.

The partnership

Both halves of this conversation belong in the operating rhythm between the business line, IT, and Finance, not in a one-time procurement review. Consider a recurring value review where you can discuss which bucket each AI investment is in, whether it has moved, or whether it now applies to multiple buckets. Pair that with a monthly cost review where you can discuss how consumption and value are tracking against the forecast, where alerts are firing, and where the kill switches (caps) are needed.

Before the next review cycle, the things below are worth doing:

1. Sort every active AI investment into the three buckets: velocity, productivity, or savings.

2. Build a consumption forecast for each AI tool, with a cost model aligned to the value creation.

3. Align the conversation with the business line and Finance.

4. Schedule transparent reviews with the business line and Finance upfront, at regular intervals, for tracking.  

Motti Finkelstein is the Founder, Chairman, and CEO of EMDO Inc, leading corporate-wide investment, advisory, incubation, and board engagement activities across global affiliates, focused on supporting next-generation market leaders through operational expertise, strategic guidance, and long-term value creation.

He previously served as Corporate Vice President and CIO at Intel Corporation, where he led global IT through the company's IDM 2.0 transformation and sponsored Intel Forge, a centralized GenAI governance platform. Earlier in his career, Motti was Chief Technology Officer for Capital Markets at BMO, a Senior Advisor at McKinsey, and held senior technology roles at Citi. Motti is a 2026 CIO Hall of Fame inductee, a 2026 National ORBIE Super Global CIO finalist, a 2024 NYCIO ORBIE Super Global Award winner, and a Forbes Technology Council contributor.