Enterprises Spend Big On GenAI But Can’t Prove it’s Working

Return on AI Institute’s Laks Srinivasan offers this commentary on how enterprises are spending big on GenAI but can’t prove it’s working. This article originally appeared in Insight Jam, an enterprise IT community that enables human conversation on AI.

On the surface, it sounds like a success story: 9 out of 10 enterprise executives report that AI is delivering value to their organizations, according to 1,000 C-suite executives surveyed in our most recent study. But if you dig one level deeper, a more troubling picture emerges. Less than half (45%) are getting what they describe as a “great deal of value.” The rest are settling for moderate returns at best. Just as worrisome, most have no method in place to measure whether their investments are truly paying off, which raises the question of whether their perception of seeing value is grounded in anything real.

What’s more, while generative AI has dominated boardroom conversations and budget allocation for the past two years, our research shows that only 9% of organizations identify generative AI as their most valuable AI type. By a wide margin (50%), analytical AI, the kind that has been quietly running credit models, demand forecasts, and fraud detection systems for decades, delivers the most value to the majority of organizations. Rule-based AI in automation software follows close behind, with 40% of organizations seeing value.

Generative AI is the hardest type of AI for which to calculate economic value, with 44% of executives citing it as the most difficult to measure. This is not because generative AI is without merit, but rather because companies are deploying it in broad, shallow ways across the workforce, giving everyone access without establishing what success looks like or how it would be tracked.

The result follows a familiar pattern. Organizations roll out AI tools widely, employees start using them, and somewhere the value is supposed to materialize. When it comes time to explain what that value was worth, no one has a confident answer.

The Measurement Problem

This value gap exists in large part because enterprises suffer from a severe measurement problem. Our six-stage AI Economic Maturity Model, built from our survey data, maps how organizations evolve in their ability to measure and report AI’s economic impact. At Stage 3, where most companies currently sit, organizations assess individual AI use cases after deployment and roughly 44% achieve high value. That sounds reasonable until you see what happens at Stage 5, where organizations formally report AI value to boards and investors. At that level, 85% achieve high value from AI. Instituting strong measurement and accountability makes an enormous difference in achieving value from AI investments.

So, what specifically makes the difference for high-performing companies? Primarily, they establish baselines before deployment, track outcomes afterward, and aggregate results across the portfolio annually. Deep involvement from senior finance managers is also important. Only 2% of organizations in our study assign the CFO responsibility for AI value, but of those that do, 76% achieve high value. By comparison, 53% of CIO and CTO-led programs deliver high value, and 32% of those led by functional leaders do so. The CFO’s involvement provides institutional credibility. When finance certifies the number, the rest of the organization believes it, and leadership can make better decisions about where to invest next.

The accountability deficit extends to the workforce in ways that carry serious consequences. Our research found that more than 60% of organizations have cut or frozen hiring because they anticipate future AI-driven productivity gains. Just 2% have made large headcount reductions directly tied to AI-related gains they’ve already achieved. That’s a thirty-to-one ratio between anticipation and evidence. If those AI productivity gains don’t materialize, companies that made significant cuts may find themselves scrambling to rehire those they laid off or, worse, unable to perform critical business functions.

To see value from AI, the path forward requires that measurement begin before deployment, that finance be a partner in certifying value, and that headcount decisions follow demonstrated productivity gains rather than precede them. The organizations pulling ahead are not necessarily those with the biggest budgets or the most aggressive deployment timelines, but the ones that treat measurement as a strategic function and resist the temptation to declare victory before the numbers are in. In the AI era, experimentation may create momentum, but measurement creates credibility.