Enterprises are pouring money into AI faster than they can prove it’s working. Gartner’s 2026 1H CIO Report found 83% of CEOs are increasing AI investment — yet 59% of AI initiatives never reach production. And 71% of the 11,000 CIOs surveyed said they struggle to prioritize AI use cases that will deliver measurable outcomes.
Now the bills are coming due. Token-based pricing means costs scale with use, and vendors have taken notice. In early July, Anthropic added cost controls to Claude Enterprise — spend alerts, model defaults and analytics that pair consumption with output — one of the first moves to tame runaway "tokenmaxxing."
But those tools show only what a company spent, not business outcomes. Tying AI adoption to "true value" is key, said Mohan Sankararaman, executive vice president and CIO of First Horizon, a regional financial services company based in Memphis. "Every AI use case should have a clear purpose and ensure that purpose has ROI."
Related:How CIOs can conquer AI model churn
To deliver ROI for AI deployments, CIOs will need to work closely with CFOs and other members of the C-suite to determine the right metrics for measuring the business outcomes of AI adoption in their organizations.
Why it’s hard to measure AI ROI
The confusion starts with treating AI like ordinary software, according to Ashwin Rangan, a veteran technology executive and board advisor who has held senior roles at Walmart and ICANN. It’s better understood as a general-purpose technology, more akin to electricity. "Just making electric power available to every worker with a wall outlet is a measure of successful deployment," he said. "But it is not a measure of effective utilization."
By that logic, the new cost dashboards are electricity meters: useful but limited. They can relay how much power was utilized from the grid, but what’s missing is a "fit-for-purpose companion meter to measure business outcomes," he added.
Most organizations still report only deployment statistics. A rare few measure actual AI outcomes, and a handful track ROI, Rangan said.
The pull toward the wrong measures is strong because the useful one is hard to implement, said Ara Kharazian, lead economist at spend-management platform Ramp. "It’s very difficult to measure AI’s impact on a business, so leaders are forced to use metrics that are available but not necessarily appropriate," he said. Tokens used, employees using AI, lines of code written — "these metrics say nothing about the quality of work being produced," he explained.
The deeper issue occurs even earlier, Sankararaman said. "The biggest risk for technology leaders is moving without purpose," he said. "Doing more AI experiments doesn’t equate to more AI value."
Related:InformationWeek Podcast: Two CTOs on managing rising AI costs
And the target keeps moving, according to Kevin Rooney, CIO of consulting firm West Monroe. "The mistake has shifted over the last six months," from hype to adoption to cost management, and now to "relying on soft productivity justification without connecting it to real business outcomes," he said.
Track AI outcomes, not just token costs
Experts say measurement work should start before any AI is switched on.
Before selecting an AI tool, executives need to identify the business workflow, explained Yogesh Joshi, senior vice president of global AI platforms at TransUnion. "We identify the job to be done, understand the people, processes and artifacts involved, and establish baseline performance metrics. Then we measure how AI changes those metrics over time." None of that requires reinventing the discipline as technology leaders have been measuring the value of tech investments for decades, he said.
At First Horizon, the right stakeholders vary by project, but the technology leader and the business owner accountable for the function always align on the yardstick before the work begins — concrete measures like time to book a loan, time to launch a product or customer experience scores. "We focus on delivering meaningful progress against those metrics before expanding our [AI] investments," Sankararaman said.
Related:How ADP’s chief AI officer keeps a tight rein on AI without holding it back
Rangan lays out a parallel sequence: benchmark the current state of a workflow, bring the CFO in early so every cost element is captured, run test cases proving the AI-assisted version is measurably better, and translate inputs and outputs into monetary terms.
Where the CFO and board stand on AI
That early CFO involvement matters, because finance leaders tend to frame AI differently than technologists do. To Hemant Kapadia, CFO at planning-software maker Anaplan, measuring AI’s value means treating it less like a software purchase than a capital-allocation decision.
"Too often, enterprises evaluate AI the same way they’d evaluate any new software purchase: ‘What did this specific tool save us?,’ and ‘Can I draw a straight line to a line item?’" Kapadia said. "You end up measuring a pilot instead of measuring a capability." The capital lens asks a bigger question: not whether a pilot saved time, but whether it delivered business outcomes like changing cost structures, speed of decision making, or an organization’s competitive position, , Kapadia said.
That makes the CIO-CFO relationship strategic rather than transactional. The CIO enables — keeping AI usable, secure and governed — while the CFO evaluates whether AI is affecting the economics of the business, Kapadia said. The catch is that the two executives must align on what a deployment is meant to change before it goes live, not after.
The people to whom technology leaders answer have already made this shift. "Boards are already there," Rangan said, crediting rooms full of seasoned CEOs, CFOs and directors. "The conversation has shifted materially toward ROI. Management teams are still not there." The result is that board members are becoming increasingly frustrated that ROI is not being demonstrated as AI accelerates, he warned.
Inside the enterprise, the pressure is concrete. In 2026, boards and c-suite executives want evidence that AI is moving from pilots to delivering ROI, Sankararaman said. They want results that demonstrate that AI is delivering business value and that technology teams are using the right governance for risk management.
Where to start measuring AI ROI
For leaders facing a climbing AI token bill and a skeptical CFO, the advice converges on a disciplined start. Sankararaman advised picking two or three "bold use cases," implementing them and ensuring they align with business outcomes.
Joshi’s advice: "Start small and go deep." Once value is proven in one high-impact workflow, it’s much easier to scale measurement across the organization, he said. And know when to walk away. "AI can’t be a hammer looking for a nail," he added. "In certain cases, the decision not to use AI could very well be the most appropriate one."
The token costs, meanwhile, will keep climbing. The IT leaders who can show what those tokens buy will keep earning the investment. Those who can only report what they spent may find their AI budgets facing the same scrutiny cloud spending once did.
How is your organization measuring whether AI investments are paying off — and what’s working? Share your thoughts with us at [email protected].
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.