Enterprises are pouring cash into AI quicker than they’ll show it is working. Gartner’s 2026 1H CIO Report discovered 83% of CEOs are growing AI funding — but 59% of AI initiatives by no means attain manufacturing. And 71% of the 11,000 CIOs surveyed stated they wrestle to prioritize AI use circumstances that can ship measurable outcomes.
Now the payments are coming due. Token-based pricing means prices scale with use, and distributors have taken discover. In early July, Anthropic added price controls to Claude Enterprise — spend alerts, mannequin defaults and analytics that pair consumption with output — one of many first strikes to tame runaway “tokenmaxxing.”
However these instruments present solely what an organization spent, not enterprise outcomes. Tying AI adoption to “true worth” is essential, stated Mohan Sankararaman, government vp and CIO of First Horizon, a regional monetary companies firm primarily based in Memphis. “Each AI use case ought to have a transparent function and be sure that function has ROI.”
To ship ROI for AI deployments, CIOs might want to work intently with CFOs and different members of the C-suite to find out the best metrics for measuring the enterprise outcomes of AI adoption of their organizations.
Why it is onerous to measure AI ROI
The confusion begins with treating AI like abnormal software program, based on Ashwin Rangan, a veteran expertise government and board advisor who has held senior roles at Walmart and ICANN. It is higher understood as a general-purpose expertise, extra akin to electrical energy. “Simply making electrical energy accessible to each employee with a wall outlet is a measure of profitable deployment,” he stated. “However it isn’t a measure of efficient utilization.”
By that logic, the brand new price dashboards are electrical energy meters: helpful however restricted. They’ll relay how a lot energy was utilized from the grid, however what’s lacking is a “fit-for-purpose companion meter to measure enterprise outcomes,” he added.
Most organizations nonetheless report solely deployment statistics. A uncommon few measure precise AI outcomes, and a handful monitor ROI, Rangan stated.
The pull towards the mistaken measures is robust as a result of the helpful one is difficult to implement, stated Ara Kharazian, lead economist at spend-management platform Ramp. “It’s totally tough to measure AI’s affect on a enterprise, so leaders are compelled to make use of metrics which are accessible however not essentially acceptable,” he stated. Tokens used, staff utilizing AI, traces of code written — “these metrics say nothing in regards to the high quality of labor being produced,” he defined.
The deeper problem happens even earlier, Sankararaman stated. “The largest danger for expertise leaders is shifting with out function,” he stated. “Doing extra AI experiments would not equate to extra AI worth.”
And the goal retains shifting, based on Kevin Rooney, CIO of consulting agency West Monroe. “The error has shifted over the past six months,” from hype to adoption to price administration, and now to “counting on gentle productiveness justification with out connecting it to actual enterprise outcomes,” he stated.
Observe AI outcomes, not simply token prices
Consultants say measurement work ought to begin earlier than any AI is switched on.
Earlier than deciding on an AI device, executives must determine the enterprise workflow, defined Yogesh Joshi, senior vp of world AI platforms at TransUnion. “We determine the job to be finished, perceive the folks, processes and artifacts concerned, and set up baseline efficiency metrics. Then we measure how AI adjustments these metrics over time.” None of that requires reinventing the self-discipline as expertise leaders have been measuring the worth of tech investments for many years, he stated.
At First Horizon, the best stakeholders range by venture, however the expertise chief and the enterprise proprietor accountable for the operate all the time align on the yardstick earlier than the work begins — concrete measures like time to e-book a mortgage, time to launch a product or buyer expertise scores. “We deal with delivering significant progress towards these metrics earlier than increasing our [AI] investments,” Sankararaman stated.
Rangan lays out a parallel sequence: benchmark the present state of a workflow, convey the CFO in early so each price factor is captured, run check circumstances proving the AI-assisted model is measurably higher, and translate inputs and outputs into financial phrases.
The place the CFO and board stand on AI
That early CFO involvement issues, as a result of finance leaders have a tendency to border AI in another way than technologists do. To Hemant Kapadia, CFO at planning-software maker Anaplan, measuring AI’s worth means treating it much less like a software program buy than a capital-allocation resolution.
“Too usually, enterprises consider AI the identical manner they’d consider any new software program buy: ‘What did this particular device save us?,’ and ‘Can I draw a straight line to a line merchandise?'” Kapadia stated. “You find yourself measuring a pilot as a substitute of measuring a functionality.” The capital lens asks a much bigger query: not whether or not a pilot saved time, however whether or not it delivered enterprise outcomes like altering price buildings, velocity of resolution making, or a company’s aggressive place, , Kapadia stated.
That makes the CIO-CFO relationship strategic moderately than transactional. The CIO permits — holding AI usable, safe and ruled — whereas the CFO evaluates whether or not AI is affecting the economics of the enterprise, Kapadia stated. The catch is that the 2 executives should align on what a deployment is supposed to alter earlier than it goes reside, not after.
The folks to whom expertise leaders reply have already made this shift. “Boards are already there,” Rangan stated, crediting rooms stuffed with seasoned CEOs, CFOs and administrators. “The dialog has shifted materially towards ROI. Administration groups are nonetheless not there.” The result’s that board members have gotten more and more pissed off that ROI will not be being demonstrated as AI accelerates, he warned.
Contained in the enterprise, the strain is concrete. In 2026, boards and c-suite executives need proof that AI is shifting from pilots to delivering ROI, Sankararaman stated. They need outcomes that exhibit that AI is delivering enterprise worth and that expertise groups are utilizing the best governance for danger administration.
The place to begin measuring AI ROI
For leaders going through a climbing AI token invoice and a skeptical CFO, the recommendation converges on a disciplined begin. Sankararaman suggested selecting two or three “daring use circumstances,” implementing them and guaranteeing they align with enterprise outcomes.
Joshi’s recommendation: “Begin small and go deep.” As soon as worth is confirmed in a single high-impact workflow, it is a lot simpler to scale measurement throughout the group, he stated. And know when to stroll away. “AI cannot be a hammer searching for a nail,” he added. “In sure circumstances, the choice to not use AI might very nicely be probably the most acceptable one.”
The token prices, in the meantime, will preserve climbing. The IT leaders who can present what these tokens purchase will preserve incomes the funding. Those that can solely report what they spent could discover their AI budgets going through the identical scrutiny cloud spending as soon as did.
How is your group measuring whether or not AI investments are paying off — and what’s working? Share your ideas with us at [email protected].
