CEOs, CIOs, CFOs, and operations leaders need to agree on what their AI investment is supposed to accomplish. That means defining where AI can strengthen your competitive position, which opportunities deserve investment, and what governance you need to manage the risks. Only then can you judge the bill against the business value you expect it to deliver.
The misdiagnosis
The pattern is consistent enough now that I can almost hear it before the call starts.
The client has deployed Copilot across the business. Engineering picked up Cursor on their own and likes it. A handful of use cases have shipped, all of them interesting and yet the adoption rate has been low. Nobody at the executive table is clear on whether and why any of it is working.
According to McKinsey’s 2025 AI report, only 39% of enterprises report EBIT impact, highlighting a significant adoption-value gap.
The instinct is to blame the tools, or the adoption curve, or, especially, the cost of the tokens. I hear from clients who have started their AI deployment: “We’ve deployed Copilot, we’re using the AI functions in all our SaaS tools, and we’re developing code with our engineering team in Cursor. But it’s all too expensive. We need to bring our usage costs down.” Note the question they’re not asking: “Is it worth it? What would be worth it?” If they never defined what value would look like, or what would drive it, they are left with only one action available: to try to constrain this vaguely helpful yet astronomically high bill into something palatable for “general uplift.”
No one ever asked which question the AI investment was supposed to answer, and certainly no one had a chance to write that down before they started spending. So there is no standard to judge the results against. You cannot fail a test you never defined, which sounds like good news right up until the second budget cycle arrives.
The market has spent two years selling executives AI answers before their teams agreed on the problem. The use case frenzy took off, the departmental demands were met, but the harder conversation about what AI investment is for gets deferred over, and over, again.
Measuring harder is not the answer
We’re deep enough into the AI era now that many of my clients are facing agreement renewals with their preferred frontier models, and the terms become expensive as the incentives designed to make the tool sticky expire. The first reflex when faced with a vague bill is to measure. Count the tokens, watch the bill, build a dashboard, put it in front of the CFO.
Do it. We’ll help you! It is useful, but it is nowhere close to sufficient.
Tokens are the atomic unit of AI. They are not the atomic unit of value.
Consumption data tells you what you spent. It does not tell you whether you should have spent it, and it is completely silent on the more expensive question of what you failed to spend it on. You can constrain your costs, and present a beautiful dashboard describing confident, well-instrumented investment in the wrong things.
The useful move is upstream of measurement. Before you can judge the spend, you have to know which question the spend was answering. In my experience there are three of them, before you get to the spend question itself. Many organizations are sitting inside all of these questions right now having only named the last one.
The four questions your AI program hinges on
Question 1. How does AI change our competitive advantage?
The symptom. In 2022 the expertise and velocity of your engineers protected you from disruption. You could design, build and ship faster than the people coming after your customers, and that advantage took years to build. Suddenly, it got cheap. Barriers to entry you have relied on for a decade are now available to a competitor with a corporate card and a small team.
Who owns it. The CEO and COO, and often the CRO, because underneath it this is a question about what you sell and to whom.
What a good answer looks like. A strategic position, before a roadmap. What is your company for when the thing you were best at has become vibe codable? If what comes out of the room is a technology plan, you still haven’t answered this. You need this to design the workforce of your future, to anchor the massive change management initiative we’re all contributing to whether we realize it or not, to choose the right technical architecture to protect and accelerate your value.
Question 2. Which AI investments should we prioritize?
The symptom. You have a rapidly multiplying collection of AI deployments. Each one was individually defensible, or it proliferated on its own through the ambition and good intent of your employees. Together they are not a strategy, and nobody in the building can tell you what is missing or where value is being left on the floor.
Who owns it. The CIO, with business counterparts who care about the big picture.
What a good answer looks like. A stated theory of where the most important value comes from alongside a ranked list with named owners. That ranking is public, referenceable, and helps you make hard decisions. A portfolio where nothing loses is just an inventory with better formatting.
Question 3. What AI governance and risk controls do we need?
The symptom. You cannot answer four basic questions about your AI systems: What goes in? What comes out? Who is using it? How much risk are we willing to accept?
Trust breaks into three separable pieces. Where does your risk tolerance actually sit, scenario by scenario, and where do the business and technical answers diverge? How does your current governance maturity compare to a framework like ISO/IEC 42001 or the NIST AI RMF, and where do you want to be in twelve months? And what do existing regulatory obligations already require, regardless of what you decide?
Who owns it. The CIO, COO and CISO together, with General Counsel in the room wherever the regulatory floor is high, because the answer touches architecture, process, risk and compliance in the same breath.
What a good answer looks like. A documented risk tolerance with the tensions still visible, a maturity baseline against a named framework, and operating model design decisions with owners attached. Not a PDF that circulates for comment and lands in a folder.
Question 4. How should we evaluate AI spending and business value?
The symptom. Your AI spend is visible and uncategorized. You can produce the invoice but it has only a vague tie to what is going well (or not) for your business goals.
This starts with understanding that different kinds of AI spend create different kinds of value. Some of what you spend is necessary and non-differentiating. It keeps the business running, your competitors have the same capability, and no customer will ever choose you because of it. For that category, the job is purely to optimize cost. Get it cheaper, then get it cheaper again, and feel nothing about it.
The rest is genuinely differentiating. It is a reason someone picks you over the alternative. For that category the job is not to optimize. It is to invest, deliberately, and to stop flinching at the bill. I was chatting to a colleague in a premier engineering firm. He said one of their engineers spent $500k in a month on AI usage. I gasped before I heard what he said next: What that engineer built will pay off 10x for the company. They knew what that investment was for, and they would spend that again and again and again for the outcomes it achieved.
Most AI cost content in the market stops at “reduce your bill.” That is a procurement answer to a strategy question. A CFO asked to bring AI spend down across the board will cut both categories at the same rate, because nobody provided a framework for decision making. The cheap capability gets slightly cheaper and the differentiating capability will starve, and your company with it.
Who owns it. The CFO, with the CIO.
What a good answer looks like. A spend baseline by provider and by user, mapped against where value is truly created. Differentiating and commodity AI spend each get named investment levels, the right cost levers, and metrics that fit the category.
Nobody answers these alone
Business, finance, operations and technology each hold a piece of the same answer.
A technology answer with no finance sponsor becomes an unbudgeted pilot. A finance answer with no operations sponsor becomes a cost cut that cripples a potentially revolutionary capability. A people answer with no technology sponsor becomes a change-management program without a concrete system to change into.
The Chief People Officer belongs in this room too and is more often the one who notices first that AI adoption is not a tool selection decision. The discrepancy between the frenzied press around AI capability and the bland adoption of these tools into the enterprise only further threatens the future of the workforce who is using them, exposing the organization to disruption from organizations who are ready to redesign the way work is delivered.
To put it bluntly: If you are reading this by yourself, and you cannot get the CFO, the COO and the CIO into a room for two hours, the work is not ready. Not because the questions are too hard. Because any answer you reach alone will end up unfunded, unstaffed or unadopted by the people who were not there.
It is much cheaper to learn that today than weeks or months into a huge project that still leaves the right question unanswered.
Find your question first
You do not need a long engagement to work out which question you are in. You need a couple of hours and the right people in the room.
That is what our AI Priorities Workshop is for. It is the shortest and cheapest thing we do, and it answers three things: who you are, what you are spending money on, and what your top priorities are. You leave with a succinct, actionable summary of opportunities and obstacles across direction, alignment, trust, adoption and orchestration foundations.
If you already know your question, skip the diagnostic and go straight to the session that matches it. Future Visioning for who we are now. A Portfolio Value Assessment for what to invest in. Governance Design for trust. Spend Rationalization for the cost versus value question.
If you can name the question, you have an AI program. If you skip it, you have an AI expense.
Need help getting to the right question for your AI program? Get in touch with the Presidio Lighthouse team today.
