Is enterprise AI delivering a return on investment?
At the beginning of 2026, Gartner said artificial intelligence would remain throughout the year in what it calls the “trough of disillusionment”: the stage at which excitement gives way to harder questions about cost, reliability and commercial value.
It was not predicting an end to AI investment. Quite the opposite. Gartner forecast that worldwide AI spending would increase by 44% to $2.52 trillion during the year. Its point was that companies would become less willing to finance ambitious standalone projects without a reasonably predictable return on investment (ROI).
“AI adoption is fundamentally shaped by the readiness of both human capital and organisational processes, not merely by financial investment,” Gartner analyst John-David Lovelock said in January.
Gartner expected enterprises increasingly to acquire AI through software suppliers they already used rather than pursue new “moonshot” projects. Companies would favour tools with more familiar costs, clearer applications and fewer integration risks.
Seven months later, the evidence presents a curious picture. Companies are putting more AI systems into production and many can identify at least some financial benefit. Yet a majority still appear unable to demonstrate that the value created exceeds what they are spending.
Two enterprise AI studies published in July capture that tension particularly well.
Enterprise AI production improves while returns stall
The Fith Annual Domino Enterprise AI Report, released on 21 July, found that 57% of enterprises said the return from AI was either level with their investment or had failed to keep pace with it. The proportion was unchanged from 2025.
At first glance, that appears to confirm Gartner’s diagnosis. Companies have invested heavily, but most are still waiting for AI to produce a return greater than the amount being spent.
What makes the finding more revealing is that technical progress has not stopped. Among the 639 senior AI leaders surveyed, 93% said their organisation had improved its ability to put AI into production, up from 88% a year earlier.
Companies are becoming better at building and deploying AI. They are not necessarily becoming better at extracting business value from it.
Domino describes this as a “last-mile gap." An AI model may be technically operational without being embedded in the decisions and processes that determine how a business performs. Forty percent of respondents said employees still relied on at least one mediated method to obtain AI-generated insights, such as requesting an analysis or waiting for a scheduled report.
That is a long way from AI working naturally inside everyday business processes. Putting a model on a server is not the same as getting useful information to a sales manager, engineer or claims handler at the moment it is needed.
The results also revealed a geographic divide. Returns were no greater than spending at 51.1% of North American companies, compared with 66.9% in the UK and 67% in continental Europe.
Domino’s sample was limited to organisations with annual revenue of at least $100 million in financial services and insurance, life sciences and the public sector. The research was conducted by BARC Research on behalf of Domino, which sells enterprise AI development and governance software. Its findings are therefore useful rather than definitive.
Even so, they indicate that the obstacles are no longer confined to model performance. Access, governance, integration and employee adoption now determine whether technical capability becomes business value.
Companies are finding pockets of AI ROI
In the last week of July, Dun & Bradstreet published what appears to be a more encouraging assessment.
Its quarterly AI Momentum Survey covers 10,000 businesses across 32 countries. More than three-quarters of respondents said their AI initiatives were generating some measurable return. Around 28% reported broad or strong returns across multiple projects.
But the largest group, representing 48% of respondents, reported only “pockets of ROI”. Some projects were working, but the benefits had not spread consistently across the organisation.
This does not necessarily contradict Domino’s finding. A company can obtain a positive return from an automated customer-service application while still spending more on its overall AI programme than it earns or saves. It might be running dozens of pilots, paying for licences that employees seldom use, hiring consultants and upgrading its data infrastructure at the same time.
One successful application proves that AI can create value. It does not prove that the complete investment is profitable.
The surveys also measure different things. Domino examined whether returns were keeping pace with investment. Dun & Bradstreet asked whether businesses could identify any measurable ROI. “Some return” and “more value than total spending” are very different thresholds.
Self-reported AI ROI adds another complication. Companies may calculate it using hours saved, tasks completed or estimated productivity improvements rather than additional revenue or reduced expenditure recorded in their accounts. Those measures can be valuable, but they are not interchangeable.
Why AI productivity does not automatically become profit
The distinction between productivity and financial return runs through much of the recent evidence.
Deloitte’s 2026 State of AI in the Enterprise, based on responses collected from 3,235 senior leaders across 24 countries between August and September 2025, found that 66% of organisations reported productivity or efficiency improvements. Forty per cent had reduced costs, but only 20% reported higher revenue.
AI can help an employee draft a document, analyse information or write code more quickly. Whether that saves the company money depends on what happens to the time released.
If employees produce more, serve additional customers or concentrate on more valuable work, the gain may eventually reach the bottom line. If workloads and processes remain unchanged, the improvement may never become a financial return.
The same problem applies to AI licences purchased across an organisation. Ten enthusiastic employees may use an assistant extensively while hundreds of occasional users generate little value. Average adoption can obscure a sharp divide between highly productive users and largely unused subscriptions.
Deloitte found that only 34% of organisations were using AI to develop new products, reinvent core processes or change their business models. Another 37% were applying it at a relatively superficial level, with little or no change to existing processes.
That may explain why productivity gains are appearing faster than revenue growth. Adding AI to a poorly designed process can make individual steps quicker without improving the process as a whole.
Data readiness becomes the AI bottleneck
Dun & Bradstreet found that only 6% of companies considered their data fully ready to support AI at scale. Nearly half described it as only partially ready.
Generative AI applications need access to accurate, current and consistent company information if they are to move beyond general assistance. Customer records may be duplicated, product information may differ across systems and important knowledge may remain buried in documents or employees’ heads.
Correcting those weaknesses is expensive. Companies must connect systems, organise access permissions, improve data quality, establish AI governance and determine who is accountable when a generated answer is wrong.
These investments may eventually produce benefits beyond AI. Better data and redesigned workflows can improve conventional software and human decision-making too. That makes it harder to isolate precisely how much value should be attributed to the AI component.
It also explains why the total cost of enterprise AI implementation is routinely underestimated. The model or software subscription is often the most visible expense, but it may represent only a small part of the work required.
AI disillusionment is not an investment retreat
Businesses are not responding to uncertain returns by abandoning AI. In May, Gartner raised its worldwide AI spending forecast to $2.59 trillion, representing 47% growth from 2025.
Most of that total does not represent ordinary companies buying AI tools. Gartner said vendors and hyperscalers continued to dominate spending, with infrastructure accounting for more than 45% of the market. It nevertheless expected enterprises to expand their use of generative AI models embedded in existing applications and deploy AI agents across multistep workflows.
Gartner’s May assessment otherwise echoed its January warning. Companies still showed limited appetite for using AI to drive disruptive change, Lovelock said, favouring tactical projects that produced incremental improvements in efficiency and productivity.
Demand for AI implementation support is also growing. French technology services group Capgemini raised its 2026 revenue-growth forecast in July after second-quarter bookings increased by 9.2% to €6.55 billion. Reuters reported that customers were spending on the core-system upgrades, data organisation, application modernisation and workflow redesign needed to support AI adoption.
Capgemini’s results show that companies continue to spend on implementing AI. They do not demonstrate that those customers have already earned a return.
That combination—rising expenditure alongside greater scepticism—is not as contradictory as it appears. The trough of disillusionment does not mean interest disappears. It marks the point at which organisations discover that a promising technology requires more work, time and complementary investment than early demonstrations suggested.
Is enterprise AI delivering ROI in 2026?
The available evidence suggests that AI is generating measurable returns for some companies and particular applications. Productivity gains are becoming common, while reductions in operating costs are appearing in selected processes.
What remains much harder is converting those gains into higher revenue, wider margins or a positive return across the complete AI programme.
The July research indicates that companies are beginning to find value, but mainly in particular functions and carefully selected applications. Far fewer have turned those successes into a repeatable company-wide capability.
AI has passed the stage at which businesses need to prove it can save someone time. The more difficult test is whether those saved hours, faster decisions and automated tasks can produce more revenue, lower costs or better margins.
Seven months into Gartner’s trough of disillusionment, there are credible signs that enterprise AI can pay. For most companies, however, the return remains a collection of promising pockets rather than a transformation visible in the accounts.
Liked this article? You can support our independent journalism via our page on Buy Me a Coffee. It helps keep MoveTheNeedle.news focused on depth, not clicks.