The New Boardroom Problem: How Do You Measure AI’s ROI?

As AI spending surges, boards are demanding proof that the investment is creating real economic value — not just running up the bill.

Executive boards are experiencing a quiet shift. Last year, the prevailing critique of AI was that it appeared everywhere except the income statement. That’s no longer true. AI has arrived on the income statement — but for most companies, it has arrived as a line-item expense.

AI spending jumped 110% in a single year, according to the ServiceNow Enterprise AI Maturity Index 2026. By 2027, AI is projected to consume more than 20% of the average IT budget. The investment is real, but the ability to account for it isn’t keeping pace.

That widening gap is fueling executive skepticism. Boards of directors aren’t just reacting to the high cost of tokens. They’re frustrated by an inability to answer a more fundamental question: Is AI generating economic value?

The organizations that can answer that question affirmatively are already pulling ahead. Most AI Pacesetters — the mature organizations our research identifies as leaders — have built implementation plans with defined metrics and timelines for AI transformation. Only 17% of other organizations have done the same.

Calculating return on investment in AI requires a departure from traditional software accounting. Rigid procurement metrics won’t get you there. What’s needed is a dynamic operational framework built for the way AI actually creates value.

The Problem of Non-Deterministic Value

The primary obstacle to measurement is that AI value is inherently non-deterministic. Conventional enterprise software operates on linear transactions: A contract is signed, a platform is deployed, a process is automated. Cause and effect are visible.

AI doesn’t work that way. Applications, particularly those using large language models or predictive analytics, operate in a gray area. The cost of an LLM that sharpens executive decision-making is known to the penny. The return resists a standard invoice.

This isn’t entirely new territory. Corporations have always struggled to isolate the exact economic output of a manager, a single software engineer, or a corporate strategist. Salaries are known, but individual contributions are understood only in aggregate.

The current environment differs not in the nature of the uncertainty but in the scale of the capital deployed. At today’s spending levels, AI demands rigorous empirical scrutiny. Vague assertions of “technological necessity” are no longer sufficient.

The Enterprise AI Maturity Index puts a number on this gap. While overall AI maturity increased in 2026, many organizations have yet to implement AI-powered workflows. The tools are arriving faster than the connective tissue needed to govern and measure them.

The fix starts at the top. Executive leadership must establish a unified corporate taxonomy — defining exactly how the enterprise quantifies both cost and value — and enforce it across all business units.

The Operational Framework

To course-correct, organizations must evaluate every AI deployment across four operational pillars:

1. Adoption: What percentage of the target workforce is actually using the tool? Low adoption doesn’t always mean the technology failed; it often means the deployment skipped the change management that makes new tools stick. Either way, you can’t extract value from a tool people aren’t using.
2. Utility: Of the people using the tool, how many are getting something out of it? Adoption tells you reach; utility tells you whether the tool is working as intended. A customer service agent who opens the AI assistant every day but ignores its suggestions isn’t a utility success story.
3. Impact: Did the business metric you were trying to move — response time, case volume, error rate — show measurable improvement? This is where most organizations fall short, not because the impact isn’t there, but because no one defined the target metric before deployment. If you didn’t establish a baseline, you can’t demonstrate change.
4. Value: What is that improvement worth in financial terms? Translating operational outcomes into financial reality requires making assumptions. Those assumptions should be explicit, documented and reviewed. Transparency is what makes the number credible.

The Moving Frontier

This measurement crisis is, at its core, a leadership failure. When data science teams and finance departments calculate ROI using conflicting methodologies, the numbers diverge — and the moment they clash, strategy loses internal credibility.

Leadership also needs to recognize that any framework built today carries a built-in expiration date. Enterprises currently dedicate significant analytical resources to calculating the precise ROI of text summarization or basic search augmentation. Within two years, that exercise will be obsolete — not because the technology loses utility, but because it becomes standard infrastructure.

Consider enterprise internet search. Modern corporations don’t calculate the ROI of a browser search query. It stopped being a competitive advantage the moment it became a baseline expectation.

As the technological frontier advances, the definition of value inevitably shifts. Leadership’s job is to build an AI scorecard — and to keep evaluating it well beyond deployment.

The leading enterprises have stopped debating whether AI works. They’re executing on specific parameters, measuring what moves, and focusing on what performs. That analytical specificity is what separates AI programs that build durable value from those that remain expensive experiments.