Editor's note:Aarif Nakhooda is the Chief Financial Officer of CoreAI and Customer Success atKeystone.ai, a San Francisco-based company providing global technology and consulting services to large enterprises, law firms, and governments. The views expressed in this article are solely those of the author.

Artificial intelligence has transformative potential across all industries. Yet, many financial leaders—including my fellow CFOs—are still grappling with a core question: How do we measure the value of AI?

By 2025, global AI spending is projected to reach unprecedented levels, reflecting rapid adoption across industries. According to IDC data, global spending on AI hardware, software, and services will exceed $337 billion in 2025. Financial leaders are deeply involved in these decisions. However, despite the spending surge, less than half of the AI projects reviewed by finance teams are considered to generate real value.

This gap between investment and perceived returns stems not from a misunderstanding of the technology, but from outdated evaluation methods.

Intangible benefits

First, many of AI's benefits are intangible. Better predictions, faster decisions, stronger customer engagement—these are undoubtedly valuable but difficult to quantify. Take a company using AI to personalize its website, for example. Users spend more time interacting with the brand. This is clearly a positive signal, but how much of that behavior translates into revenue? And how much interaction would have happened anyway?

Second, AI operates in complex, multi-factor environments, making attribution difficult. Suppose we deploy a machine learning pricing algorithm and observe a sales increase. Is this due to AI, or to a concurrent marketing campaign, seasonal peak, or pent-up demand for restocked items? This problem has long existed in marketing, and is now spreading across the entire operation. Isolating AI's impact is rarely straightforward.

Third—and most importantly—AI systems continuously improve over time. Unlike depreciating capital equipment, AI models learn and add value. Traditional ROI assumes linear returns over a fixed period: invest $10 today, earn $20 in a year. AI does not follow this trajectory. Returns may be modest at first, but grow exponentially as the system adapts. Capturing this non-linearity in traditional financial models is extremely difficult.

Rethinking the value of AI

None of this means we should abandon financial discipline. But it does suggest that we need to improve how we measure value. More forward-looking CFOs are beginning to combine traditional ROI with broader strategic assessments—considering intangibles, uncertainty, potential upside, scalability, and long-term learning curves.

Some are borrowing frameworks from venture capital or innovation finance, which tolerate early ambiguity in pursuit of long-term gains. Others are developing new internal metrics: for example, "AI-adjusted ROI," or performance dashboards reflecting learning progress, accuracy improvements, and decision speed. These tools align more closely with the nature of AI—and ultimately, more honestly reflect what success looks like.

As AI continues to evolve into a general-purpose technology (similar to electricity or the internet), CFOs must shift from financial gatekeepers to strategic enablers. Rather than viewing AI merely as a cost, we should see it as a capability: one that unfolds over time, compounds with use, and resists simple quantification.

The better we become at evaluating AI on its own terms, the faster we can unlock its true value. This shift requires not just new models, but new thinking.

There is a famous anecdote in executive circles: A CFO asks, "What if we invest in our employees and they leave?" The CEO replies, "What if we don't invest, and they stay?"

I believe AI now requires the same reframing. Yes, we should ask, "How much will this cost?" But we must also ask, "What is the cost of waiting?"

In a world where AI is rapidly reshaping competitive advantage, the greatest risk may lie not in making the investment—but in missing it.

Editor's note: This article is the first in a two-part series.