Enterprise spending on generative AI has surged over the past year, but for many CIOs, the hardest conversations are only now beginning. Boards and CFOs are no longer asking whether the organization is investing in AI. They are asking what it’s getting back — in measurable financial terms.
According to analysts at Forrester Research, genAI budgets have increased substantially year over year, yet a majority of organizations still struggle to demonstrate sustained return on investment. Early pilots often look promising, but value becomes harder to explain as systems scale, costs fluctuate, and governance expectations rise.
Interviews with analysts, CIOs, and AI platform and governance leaders point to a consistent pattern. The problem is not that AI fails technically. It’s that enterprises are applying legacy budgeting, operating, and accountability models to a technology whose economics behave very differently. As a result, ROI erodes not because AI stops working, but because organizations lose the ability to explain, defend, and prioritize it.
Analyst framing: from cost control to value co-creation
From an analyst perspective, the AI ROI debate is best understood as part of a broader convergence between IT and finance. Greg Zorella, lead principal analyst at Forrester covering IT financial management, argues that high-performing IT organizations no longer treat finance as a gatekeeper focused on cost containment. Instead, IT finance becomes a capability for strategic value delivery — connecting technology investment directly to business growth and competitive advantage.
“IT finance isn’t there because IT spends a lot of money,” Zorella said. “It’s there because IT spend can really drive strategic outcomes for the enterprise.” That distinction matters for AI. Traditional IT investments — ERP systems, infrastructure refreshes, SaaS licenses — fit relatively well into established financial models. Generative AI does not. Costs are consumption-based, usage patterns are unpredictable, and benefits are often indirect or risk-adjusted rather than transactional.
Zorella notes that many enterprises intellectually recognize this shift but underestimate the organizational lift required to act on it. Mature cost transparency depends on shared attribution models, reliable data, and agreement across IT, product, sales, and marketing about how value is defined.
“Trying to do that all at once is just too much,” he said. The organizations making progress tend to start with narrow proof points that demonstrate how better financial visibility improves decision-making.
Importantly, Zorella challenges the assumption that exceeding IT budgets is inherently negative. Overspending may be rational — if it reflects deliberate investment in higher-value initiatives. The real failure, he argues, is overspending without a prioritization mechanism that allows leaders to deprioritize lower-impact work when new opportunities emerge.
CIO decision reality: budgets don’t expand forever
That analytical framing meets a far more constrained reality inside the enterprise. Sumit Johar, CIO of BlackLine, which makes finance automation and management software, describes AI investment moving through a familiar cycle. In recent years, initial skepticism gave way to peer pressure as boards and executives demanded visible AI initiatives. Today, that phase is ending. Finance leaders are asking harder questions, and AI is no longer treated as a special category exempt from scrutiny.
“If I tell my CFO that 95% of employees are using AI, that doesn’t mean anything,” Johar said. “It’s like saying 100% of employees use email. Finance cares about impact on profitability, revenue, or risk — everything else falls flat.”
Johar draws a sharp distinction between two classes of AI investment. The first is broad productivity platforms — what he calls “everyday AI” — that help employees write, search, summarize, or analyze information. These tools can be transformative culturally, but they are notoriously difficult to quantify. Engagement metrics and self-reported productivity gains rarely survive financial scrutiny.
The second class consists of outcome-driven AI initiatives tied explicitly to business priorities: accelerating customer onboarding, reducing deployment time, lowering operating costs, or increasing the revenue pipeline. These initiatives compete directly with other enterprise investments and are evaluated accordingly.
What has changed most, Johar says, is that AI spending is no longer additive. CIOs are not receiving incremental budget increases “because AI.” Any additional investment must be funded by reallocating existing budgets. “Nobody is blindly throwing money at AI anymore,” he said. “If we want to spend more, we have to move things around.”
At BlackLine, AI governance reflects that reality. Proposed initiatives are reviewed jointly by IT, finance, and business leaders, with explicit expectations for outcomes and accountability. The goal is not to slow experimentation, but to ensure that responsibility for value creation does not sit solely with the CIO.
“This is a business transformation problem, not a technology problem,” Johar said. “If ownership stays only with IT, you’ll never get the value you’re expecting.”
Operational failure modes: why ROI collapses at scale
Even when AI initiatives clear budget hurdles, many fail to deliver sustained ROI once they move beyond pilots. According to Jim Olsen, CTO of AI lifecycle management and governance platform maker ModelOp, the breakdown is rarely caused by a single flaw. It is structural. Early AI projects are typically developed in controlled environments with limited data and predictable usage. Costs appear manageable, and performance looks strong. Production environments behave very differently.
“You develop something locally and it looks very doable,” Olsen said. “But once it hits production, usage patterns change, contexts explode, and suddenly the true cost shows up.”
Generative AI amplifies this problem. Free-form user interaction increases token consumption unpredictably. Models are embedded across workflows and reused by multiple teams, making it difficult to attribute cost or value to specific outcomes. Without clear inventory and lifecycle tracking, enterprises end up managing AI spend in aggregate — while value is created or lost at the margins.
Olsen says many organizations lack even a basic understanding of what AI systems they have in production. “If you don’t know what’s out there, you can’t measure it, govern it, or tie it back to ROI,” he said.
The result is a familiar pattern: promising pilots followed by cost overruns, followed by skepticism. In some cases, high-profile missteps make organizations risk-averse, slowing future adoption even where AI could deliver real advantage.
The remedy, Olsen argues, is to treat AI as industrial infrastructure rather than experimental tooling. Lifecycle management — covering development, deployment, monitoring, and retirement — is not bureaucratic overhead. It is the only way to maintain accountability as models evolve and usage grows.
Governance and defensibility: when value must be proven
Operational discipline alone, however, is not enough. As AI investments face regulatory and board-level scrutiny, governance increasingly determines whether ROI can be defended at all. Anthony Habayeb, CEO and co-founder of AI governance software vendor Monitaur, argues that many AI initiatives fail under review not because they perform poorly, but because success was never clearly defined.
“We’re running around with a hammer looking for a nail,” he said. “If you don’t know what success looks like at inception, you can’t defend ROI later.”
Governance failures often surface only after deployment, when organizations attempt to retroactively justify spend. At that point, gaps in documentation, monitoring, and accountability become liabilities. Projects that lack clearly articulated objectives or outcomes are easy targets when budgets tighten.
Habayeb challenges the idea that governance is primarily about compliance. In practice, he says, governance improves ROI by exposing unknown risks and optimization opportunities. As organizations introduce structured validation and monitoring, they often identify ways to improve accuracy, robustness, and efficiency — directly enhancing business impact.
Regulatory pressure is accelerating this shift. Frameworks such as the EU AI Act are pushing organizations to formalize oversight, but Habayeb says the smartest enterprises are using regulation as a forcing function to build broader governance capabilities.
“Governance shouldn’t be a separate compliance line item,” he said. “It should be part of how you make AI work for the business.”
From enthusiasm to endurance
Taken together, these perspectives point to a maturing phase of enterprise AI adoption. The question is no longer whether AI can deliver value, but whether organizations can prove that it does — consistently, transparently, and under scrutiny.
The enterprises making progress share several traits. They align AI investment with business strategy rather than treating it as a standalone category. They build financial models that accommodate consumption-based costs and indirect value. They enforce operational discipline across the AI lifecycle. And they embed governance early — not as a brake on innovation, but as a foundation for trust and sustainability.
For CIOs planning 2026 budgets, the message is sobering but constructive. AI will not justify itself. Value must be designed, measured, and defended, using tools and practices that many organizations are only now beginning to develop.
The era of AI as an experiment is ending. The era of AI as an accountable enterprise asset has begun.