Closing the AI Value Gap with Strategic Expenditure Discipline

Most capital decisions are easy to see coming. AI is not. It enters through dozens of small, separate choices, none of which looks big enough to need a business case, until together they reshape what you spend and what you get back.

A copilot license in one team. An AI add-on bolted onto a platform finance already pays for. A token-billed workflow no one outside the team can see. A promising pilot in operations that never makes it past the department door. Each one is small enough to fit inside an operating budget. Put together, they may be one of the largest, and least governed, capital decisions your organization makes this year.

Those scattered decisions are where the AI value gap opens up: the distance between the value you expect from AI and the value you actually keep. Most leaders blame technology. The model was not ready. The data was messy. People did not adopt it. The use case was vague.

In most organizations, though, the technology works well enough. What fails is the way AI expenditure gets evaluated, delayed, scattered, or blocked before it can become enterprise value.

Closing the AI value gap with strategic expenditure discipline starts with a shift in how you see the spend. AI is no longer an IT cost or a productivity experiment. It changes how the business operates, how customers are served, how decisions get made, and how you compete. That makes it strategic. And strategic expenditure needs more than enthusiasm on one side and control on the other.

The AI Value Gap is not just a Technology Problem

The AI value gap is the distance between what organizations expect from AI projects and what they actually capture and keep, after the cost of integration, governance, risk, and the value that leaks to vendors. It is as much a gap in investment discipline as in technology maturity. The major studies all point the same way.

Boston Consulting Group, which popularized the term, finds that only about 5% of companies are generating outsized value from AI: the “future-built” few who achieve roughly five times the revenue growth and three times the cost reduction of their peers. Close to 60% report little material value despite real investment. Stanford’s 2026 AI Index describes the same pattern as a widening gap between what AI can do and how prepared organizations are to manage it. McKinsey finds that most firms have yet to see enterprise-wide bottom-line impact from generative AI, and that fewer than one in five track KPIs for it at all.

The gap shows up in familiar ways. Tools get bought but never embedded in meaningful work. Productivity gains never convert into business outcomes. Pilots stay isolated. Lessons get lost between teams. Spend grows without visibility. And the best ideas are stopped before they reach the person who can make the strategic call.

This is why buying more tools will not close the gap. More tools increase activity without guaranteeing value. Most people are already using AI. What matters is whether that use is becoming measurable, scalable, and worth funding. It is also why a growing number of teams now argue that AI projects should be managed as strategic expenditure, not routine IT spend.

Where AI Value Gets Stuck

AI value usually starts closest to the work, where people can see what is slow, manual, or duplicated. But the moment an idea needs funding, platform access, data, a security sign-off, or a business case, it hits two functions built to slow things down:

• IT, trained to protect systems, data, architecture, and security.
• Finance, trained to protect budgets, predictability, governance, and return.

Both are right, and neither is enough alone. The value gap widens when fast-moving operational opportunity is forced through decision models built for a slower, more predictable technology cycle. The answer is to keep their rigor while updating the assumptions behind it.

IT is Right to Worry, but “no” Creates its Own Risk

IT has every reason to be cautious. AI raises real questions about data privacy, security, architecture, model governance, vendor risk, integration, and long-term support. No CIO wants the organization littered with unsupported tools, unapproved data flows, or clever internal apps that nobody can maintain once the person who built them moves on. The instinct to control is rational.

But a blanket “not yet” does not stop AI adoption. It pushes it underground. Employees who cannot reach approved tools open personal accounts and free plans, putting company data into systems IT cannot see and cannot govern.

Here is how it plays out. A marketer builds a maturity-assessment widget for the website. Someone in delivery ships a project-portfolio app and then moves on. A year later, IT inherits a pile of unsupported apps running on private accounts, plus a bill to standardize them. By focusing only on the risk of adopting AI, IT ends up creating the risk of shadow AI: privacy exposure, no observability, fragmented tooling, and orphaned apps no one can support.

The caution is earned. Zillow wound down its home-buying business after model error met balance-sheet exposure, taking an inventory write-down of roughly US$304 million. A tribunal held Air Canada responsible for inaccurate information its chatbot gave a customer. Uncontrolled automation at the front end can erase the savings it appeared to create, which is exactly why governance belongs inside the business case.

The fix is governed experimentation. Right now, leaders are choosing between experimentation they can see and experimentation they cannot.

Finance Wants Proof, but AI Value Often Compounds Before it Shows Up

Finance is right to be cautious too because AI spend is moving from negligible to material faster than most budgets can track. What starts at twenty dollars a month becomes two thousand, then ten thousand, then, for a heavily augmented role, a meaningful share of a salary, because if a tool genuinely multiplies someone’s output, the return can justify it.

For enterprise buyers, AI costs rarely arrive as a single price rise. They show up in how technology is bought and consumed: usage-based pricing, premium AI features, cloud consumption, vendor add-ons, subscriptions, and token-based services. Spread across a year, routine OpEx can become a strategic cost base quickly.

So, AI needs financial discipline. It needs business cases, prioritization, and governance. But if Finance applies old ROI tests too early, some of the best opportunities never survive the first approval gate. AI value is often uncertain at the start and compounding over time. The first use case may not show the full return. The first pilot is usually more about learning than payback. The first workflow may create reusable capability for five others. Techniques such as AI-adjusted discount rates can price that uncertainty into the evaluation rather than penalizing it with a blanket rejection.

Economists have a name for this shape. Brynjolfsson, Rock, and Syverson describe a productivity J-curve, where early returns look weak precisely because the complementary investment (data, integration, redesign, training) is front-loaded. The implication for capital planning is straightforward: traditional investment logic rewards predictable returns, while AI tends to create value through iteration, reuse, and speed.

None of this means loosening control. It means building a discipline for uncertainty: one that prices not only the direct cost of acting, but the opportunity cost of not acting: the customers, products, and people you cannot see because you lack the analytic bandwidth a better-equipped competitor already has. This is where strategic budgeting connects to strategic expenditure: the budget stops being a ceiling and becomes a way to fund uncertainty on purpose.

Scattered Spend Creates Scattered Value

The most dangerous AI spend is rarely the big transformation program with executive attention. It is the long tail of small, scattered buys no one can see clearly, each harmless on its own but together scattering both the spend and the value.

The organization spends the money, but the learning stays local. Skills stay scattered. Teams duplicate effort. Good ideas stay trapped in departments. Successful pilots never become shared capability. Deloitte finds most leaders expect fewer than 30% of their AI experiments to reach full scale within six months. The result is activity, not transformation.

Strategic expenditure discipline changes that. Capture AI demand in one place, and initiatives can be compared, platforms shared, learning compounded, and winners scaled, the same portfolio view executives already apply to strategic expenditure across the capital plan.

AI May be OpEx, but the Decision is Strategic

Part of why AI is hard to govern is that it does not enter the business through the usual investment channels. It rarely looks like a major capital project. It does not trigger the same approval pathway as a plant upgrade, a system implementation, or an infrastructure program. It arrives incrementally, through teams, tools, vendors, and operating budgets, so most commercial AI consumption lands as operational expenditure, which, once it materially shapes the business, becomes strategic operational expenditure rather than routine running cost.

But the strategic significance of AI should not be judged by where the cost sits in the ledger. AI expenditure becomes strategic when it changes how the organization works, competes, serves customers, builds capability, or protects future relevance. The accounting treatment is operational. The business impact is strategic.

That is the distinction many organizations miss. Strategic expenditure = CapEx + strategically significant OpEx. AI deserves CapEx-grade rigor whenever it materially affects future business performance, which, in asset-intensive industries where every capital decision already faces scrutiny, should feel familiar. The ledger may call it OpEx. The board should not treat it as routine.

The Risk of Doing Nothing is no Longer Zero

Most investment processes are built around the risk of action. What if the project fails? What if the cost blows out? What if the vendor disappoints? What if adoption is poor? What if the business case is wrong? These are legitimate questions.

But AI has made the other side of risk harder to ignore. What if the organization does not move? What if competitors reduce cost-to-serve faster? What if customers expect smarter digital experiences? What if employees grow frustrated with outdated tools? What if capability gaps compound for years before anyone notices?

When decision-makers model only the risk of action, AI initiatives look too uncertain to approve. In a fast-moving market, inaction is itself a decision, and it carries its own consequences.

In AI, “not yet” can be the more expensive decision. For asset-intensive businesses, inaction may be the most serious blunder on the board’s table.

AI Cannot Sit with IT or Finance Alone

IT and Finance are essential, but AI value crosses the whole organization: operating models, customer experience, product design, productivity, and competitive position. That makes it an executive issue, not a departmental one.

Responsibility does not rest with the CIO or the CFO alone. It rests with the CEO and leadership team, the only people who can weigh the competing pulls in one frame: IT’s caution on security, Finance’s caution on cost, Operations’ push for speed, and the business’s fear of being left behind.

AI does not promise a safe, predictable return. It promises a risky, potentially large one, and that call belongs at the top. AI value crosses functional boundaries, so AI governance has to cross them too.

How Strategic Expenditure Discipline Turns AI Activity into AI Value Capture

Strategic expenditure discipline speeds the right initiatives up rather than slowing AI down. It creates a way to see demand, compare opportunities, evaluate risk, prioritize funding, and scale what works. It gives IT a structured role in managing risk, Finance a better way to evaluate uncertainty, Operations a route for high-value ideas to be seen, and executives a portfolio view. That is how AI activity becomes AI value capture.

1. Capture AI demand in one place. Replace disconnected pilots, hidden subscriptions, and duplicated experiments with a single view of where AI interest is emerging and where investment may be needed.
2. Start with the business problem, not the technology. Do not take AI and hunt for a use case. Take the business problem and ask whether AI is part of the best solution, which is really a question of how to choose your AI projects. In practice, AI becomes a dimension of almost every business case (sustenance, replacement, growth, compliance) rather than a separate category of investment. One option simply uses more AI than another.
3. Classify AI across CapEx and OpEx. Recognize that strategic AI expenditure may include platforms, integration, data infrastructure, subscriptions, token usage, cloud consumption, copilots, and internal capability. This stops material AI spend from hiding in disconnected operating budgets.
4. Evaluate two-sided risk. Assess both the risk of acting and the risk of inaction, so decision-makers see uncertainty, opportunity cost, and strategic exposure in the same frame.
5. Prioritize by strategic value, not just short-term ROI. Weigh financial and non-financial value together: productivity, revenue potential, customer value, speed, risk reduction, capability building, strategic alignment, and competitive relevance.
6. Fund concentrated bets, not endless pilots. Avoid spreading investment too thinly, and weigh platform investments against point solutions. OECD evidence supports this: 87% of firms putting more than 30% of their R&D into AI treat it as critical to core processes, against just 38% of those spending under 10%. The discipline to take from that is simple: back fewer, bigger AI bets and resource them properly.
7. Scale what works. Move beyond isolated pilots into shared platforms, reusable patterns, and governed adoption pathways. The pattern that separates value from activity is embedding: Morgan Stanley built its AI assistant into advisor workflows and reached around 98% adoption, while Klarna built its assistant into frontline service, where it handled two-thirds of chats within a month. The value came from AI living inside the workflow, not bolted on beside it.

The aim is simple: stop good AI ideas from disappearing before anyone can act on them, without burying them in process.

From Clever Pilots to Compounding Capability

Many organizations still measure AI maturity by activity. How many tools are in use? How many pilots are underway? How many people have access? How many tokens are we consuming? Those questions matter, but they are not the test.

The better test is whether AI capability is compounding. Are teams learning from each other? Are successful use cases being scaled? Are data, platforms, and governance improving with each wave of investment? Is AI changing the operating model, or only speeding up individual tasks?

There is a bigger prize than efficiency. Use AI to do the same work faster and you take out cost. Use it to do what was not possible before and you create new value. This is the distinction Ethan Mollick describes as automation versus augmentation. The higher rungs of the maturity path are where that shift happens.

A practical maturity path looks like this:

Level 1: Individuals use AI informally.

Level 2: Departments experiment with AI.

Level 3: Teams use approved tools.

Level 4: AI initiatives are captured and compared centrally.

Level 5: AI is prioritized through strategic expenditure governance.

Level 6: AI improves productivity, cost, speed, and customer experience.

Level 7: AI transforms products, services, or business models.

The higher the maturity level, the greater the potential value and the greater the need for visibility, prioritization, and executive decision-making. The measure of AI maturity is how much of that experimentation becomes reusable business capability.

The Winners Will not just Spend More on AI

The AI value gap will not close by buying more tools, running more pilots, or telling every team to experiment harder. That raises AI activity without raising AI value capture.

Remember where this started: a scatter of small AI buys, each one small enough to fit inside an operating budget, now adds up to one of the largest and least governed decisions on your books. You cannot cross new territory with maps drawn for slower, more predictable technology. The organizations that close the gap will treat AI as strategic expenditure: visible, comparable, prioritized, and actively governed across CapEx and OpEx.

Done well, this gives the strongest AI initiatives a clearer path: better business cases, sharper risk evaluation, real executive visibility, and tighter alignment to strategy. It is the opposite of red tape.

AI spending is already happening. The only question is whether it produces scattered productivity gains or compounding strategic value. The winners will not be whoever spends the most on AI. They will be whoever is disciplined enough to turn that spending into lasting advantage.

See your AI spend as a portfolio

Stratex Online gives executives, CFOs, and capital planners a single, governed view of strategic expenditure across CapEx and OpEx, so the right AI initiatives get seen, compared, prioritized, and scaled. Start with AI projects as strategic expenditure, explore the broader strategic expenditure framework, or see how it applies to AI in CapEx management.

FAQ on the AI Value Gap