CEOs keep pouring cash into AI. Returns? Not always clear. CFOs stand guard over the checkbook, caught between ambition and accountability. Amy Butte, a four-time CFO who led finance at the New York Stock Exchange and most recently Navan, puts it bluntly. “It’s always incumbent on a CFO to measure outcomes or measure impact and to translate the language of numbers into behavior or behavior into the language of numbers,” she told Business Insider. But today’s pace demands more. Rapid AI shifts mean new metrics. CFOs aren’t natural risk-takers. They must adapt anyway.
Butte lays out a straightforward path. First, pin down success. What do investors want? Revenue growth. Pretax earnings. Return on equity. Align every AI dollar there. Don’t chase shiny tools that miss the mark. Second, track the gears turning underneath. Compare AI-handled customer support against human efforts. Gauge code delivery success rates within deadlines. These behind-the-scenes numbers feed the big goals. Third, get in the game. “This is not the time for finance leaders to sit on the sidelines,” Butte says. “It’s important to take risks. It’s important to try new things in an environment where change really can move the needle.”
Measurements evolve. Teams must sync on changes. Butte warns against fool’s gold, like fixating on steps walked while ignoring diet in a weight-loss push. Businesses do the same with leaderboards or tokenmaxxing—obsessing over AI token spend without fixing core flaws. “Just saying it is one thing. Measuring it—and measuring it with the right things, and sometimes it’s multiple things—is critical,” she adds. “And to me, that’s the responsibility of a CFO. Tokenmaxxing isn’t going to solve for a bad product.”
Her words hit harder amid the spending frenzy. Goldman Sachs reports companies blowing past AI inference budgets, with those costs now nearing 10% of headcount expenses and on track to match salaries soon, as shared in recent X discussions citing the bank’s analysis. KPMG’s Global AI Pulse survey of over 2,100 leaders shows U.S. firms planning $178 million average AI outlays over the next year. Asia-Pacific budgets top $245 million. EMEA trails at $157 million. Worldwide IT spend? Headed to $6.31 trillion in 2026, up 13.5%, fueled by AI infrastructure and cloud, per Forbes.
Compute costs already outstrip people in some AI-heavy outfits. Nvidia’s Bryan Catanzaro notes, “The costs of compute have exceeded the costs of my people.” Uber’s CTO burned the full 2026 AI budget by March. Swan AI exhausted two months’ allocation in weeks, its CEO boasting of scaling via intelligence over headcount. Enterprises dropped $4 billion on AI coding tools last year—more than IT, marketing, customer success, and HR AI budgets combined. Yet over half of early 2026 GitHub commits were AI-generated. Problem? Bugs galore. Developers say AI code takes longer to debug. A METR study: AI users 19% slower overall. Measure lines of code? Wrong yardstick. Track bug resolution time instead, urges Forbes Tech Council.
OpenAI embodies the tension. CEO Sam Altman eyes a Q4 IPO despite $200 billion projected burn before profits. CFO Sarah Friar flags risks in the spend plans, deeming the firm unready, reports Forbes. Tokenmaxxing thrives elsewhere. Databricks, Nvidia, Meta cheer engineers dropping thousands on tokens. Sendbird’s leaderboard crowns top spenders “AI gods.” Nvidia’s Jensen Huang: “If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed.”
Levi Strauss offers a counterpoint. Partnering with Wipro, it swapped manual spreadsheets for machine learning in revenue and earnings forecasts. Results: sharper predictions. Less team ping-pong. Staff freed for why numbers shift, not just crunching them. AI augments FP&A, doesn’t replace it. Humans still probe model outputs amid business swings, per Qubit Capital.
So budgets balloon. Proof lags. Quarterly calls will force the issue. Savings in ops. Efficiency jumps. Faster market entry. Higher output per head. New AI revenue streams. Tie spend to these, or face the squeeze. Early movers pay premium as AI labs hike prices. But value demands discipline. Human talent pairs with compute. Biggest checkbooks won’t win. Measurable wins will.
CFOs throttle costs early. Old methods work if tuned right. Butte’s playbook fits: define. Track. Engage. Ignore token races. Fix products first. As inference eats headcount dollars, and $178 million averages loom, finance chiefs decide. Fund the future. Or fund the hype.
WebProNews is an iEntry Publication