ROI Wasn’t Built in a Day. Are we measuring what counts with AI?

When ChatGPT arrived in November 2022, the economic expectations surrounding generative AI were extraordinary.

In 2023, Goldman Sachs projected that generative AI could increase global GDP by 7% over a decade. McKinsey estimated between $2.6 trillion and $4.4 trillion in potential annual economic value across 63 use cases.

These were forecasts of economic potential, not promises of financial returns. They nevertheless helped establish an expectation that a technology capable of transforming work would also deliver substantial economic value.

Four years later, we have considerably more evidence about what generative AI can do. Whether those capabilities are delivering the returns organisations anticipated is a more complicated question.

Perhaps the complication begins with what we expected those returns to look like.

Traditionally, return on investment (ROI) evaluates financial gains relative to the cost of an investment. Reduce expenditure, increase revenue or improve profitability, and we have familiar ways to assess the result.

AI investments, however, are intended to create different kinds of value, not all of which appear immediately on a financial statement.

Consider customer service.

Research published in the Quarterly Journal of Economics examined an AI assistant used by 5,172 customer-support agents. The researchers found a 15% average increase in issues resolved per hour, with considerably larger improvements among less-experienced workers.

Efficiency was only part of the story. Customers also became more polite and less likely to request a manager.

Now consider innovation.

In an experiment involving 776 professionals at Procter & Gamble, individuals working with AI produced proposed product solutions rated as highly as those developed by two-person teams without AI. The AI-assisted participants also generated solutions that better integrated technical and commercial perspectives.

Neither study establishes the financial return on these investments. Together, however, they illustrate different forms of potential value: operational efficiency, customer experience and innovation.

That distinction matters.

Some investments are intended to reduce the cost of existing work. Others aim to improve services, strengthen decision-making, accelerate innovation or create new commercial opportunities.

If the intended outcomes differ, should we expect to evaluate every investment using the same measures and timeframe?

Recent enterprise research makes this question particularly interesting.

McKinsey's 2026 global AI survey found that 80% of respondents reported improvements in individual productivity, while 50% reported better decision-making. Yet only 37% attributed any positive contribution to organisational earnings before interest and taxes (EBIT).

PwC's 2026 survey of 4,454 CEOs found that 56% reported no significant financial benefit from AI, while 12% reported both increased revenue and reduced costs.

These are self-reported findings about AI broadly, not audited financial returns from generative AI alone. They measure different outcomes across different samples.

Nevertheless, they reveal an important tension.

Creating an improvement is not necessarily the same as capturing a financial return from it.

A task might become faster without reducing the cost of an entire workflow. Better customer experiences may build loyalty without immediately increasing revenue. Innovation may generate promising solutions long before they become commercially successful products.

Even when value is created, establishing how much can be attributed to AI rather than other organisational changes is difficult.

The investment itself also extends beyond purchasing the technology. Integration, data infrastructure, training, verification, governance and ongoing operating costs all matter when calculating financial returns.

Then there is time.

Deloitte's 2025 research found that most surveyed executives reported taking two to four years to achieve satisfactory ROI on a typical AI use case. Only 6% reported achieving a return within the first year.

That raises another question: When should we reasonably expect an AI investment to pay off?

An investment intended to reduce administrative costs might reasonably be expected to deliver measurable benefits relatively quickly. Developing a new product, entering a market or building a strategic capability may require a longer horizon.

A longer timeframe does not automatically make an investment worthwhile. Future opportunities remain uncertain, and anticipated value still needs to be tested against evidence of progress.

This brings us back to a question that should be asked before an AI investment is made: what would a worthwhile return actually look like?

Financial returns remain essential. Organisations need to understand what they are spending, what they are earning or saving, and whether the benefits justify the full investment.

Financial ROI is not interchangeable with every operational or strategic benefit. Nor does the absence of an immediate financial return necessarily establish that an investment has failed to achieve its intended purpose.

For leaders, that means being clear about what an investment is intended to achieve, what evidence would show progress and when that value should reasonably emerge.

An investment intended to reduce costs should ultimately demonstrate savings. One intended to improve customer experience should be evaluated against service outcomes and their broader commercial implications. An innovation investment may require evidence of progress before its eventual financial value becomes clear.

Four years after ChatGPT's arrival, the research offers neither a universal success story nor a universal disappointment. It shows meaningful benefits, uneven financial outcomes and considerable uncertainty about how value translates into returns.

Perhaps the question is not simply whether AI is delivering ROI.

Are we evaluating AI investments against the outcomes we intended to create, or against the financial returns we expect to see immediately?