Enterprise AI ROI: the 2026 guide.

01What the research says about AI returns in 2026.

Spending is rising faster than proof. Gartner forecast in September 2026 that worldwide AI spending would grow 49.5 percent in 2026. BCG found that companies' AI spending rose from about 1.7 percent of revenue in late 2025 to 3.3 percent in 2026, and that more than 80 percent of it now sits outside the IT budget.

Returns are rising more slowly. In Gartner's survey of 2,002 CIOs, published in October 2026, 13 percent said their enterprise had realised significant or transformational value from AI. In PwC's survey of 4,454 CEOs, 56 percent had seen no significant financial benefit to date, and 12 percent reported both lower costs and higher revenue. BCG's 2026 index is the most positive: almost half of the 1,330 companies it surveyed now generate value from AI, up from a small minority a year earlier.

TakeawayAI spending is growing by half in a year, and the share of leaders who can show a financial return is still a minority.

02Why AI ROI is hard to prove.

Three problems recur across the surveys. The cost side is incomplete: Gartner found that 11 percent of functions did not know what they spent on AI in 2025, and BCG's finding that most AI spend sits outside IT means much of it never reaches one ledger. The benefit side is diffuse: time saved by individuals rarely shows up as a line in the accounts. And the timing is long: Deloitte found that most organisations expect a typical AI use case to pay back in two to four years, against seven to twelve months for other technology.

The cost side matters most, because it is the one an organisation can fix first. KPMG's Global AI Pulse for the second quarter of 2026 found that organisations with full visibility of their AI operating costs report established ROI at five times the rate of those without.

TakeawayAI ROI starts with knowing what AI costs, because a return can only be proved on spend you can see.

03How to measure AI ROI.

The method is the one finance already uses for any investment, applied per workload. Count every cost: licences, model usage, cloud, build and the people who run it. Set a baseline before the work changes. Agree which benefits count, in hours, money or risk, and who signs them off. Then track value against cost each month, and decide whether to scale, fix or stop.

Gartner advised CFOs in March 2026 to treat AI as a portfolio of investments with different cost and value profiles, measured with a balanced set of indicators. Its September survey found that organisations that track returns continuously, manage AI as a portfolio and cut underperformers saw positive returns on 81 percent of their AI initiatives.

TakeawayMeasure each AI workload the way finance measures any investment: full cost, a baseline, agreed benefits and a decision at the end.

04Why generative and agentic AI are measured differently.

A chat assistant costs a seat a month, and its value is time saved by the person using it. An agent runs on usage, makes many model calls for one task, and its value is work completed. Deloitte found that 85 percent of the organisations it calls AI ROI leaders use different frameworks or timeframes for the two. Gartner expects inference cost per agentic workflow to rise more than fivefold through 2028 even as the price per token falls, so agent costs need watching per workflow.

TakeawaySeats and agents need separate measures, because one costs a fixed fee for a person's time and the other a variable bill for work done.

05What the organisations with returns do differently.

• They see the cost. KPMG found established ROI at 15 percent with full cost visibility and 3 percent without.
• They name owners. KPMG found organisations with clear executive accountability for AI outcomes report established ROI at more than three times the rate of those without.
• They run a management layer. In KPMG's third-quarter survey, 86 percent of organisations with established ROI operate a formal cross-functional or enterprise-wide AI management layer, against 31 percent of those still experimenting.
• They stop what does not pay. Gartner found organisations with rules for stopping pilots were 2.3 times more likely to be among its top performers.

TakeawayThe organisations reporting returns see the cost, name an owner, manage AI centrally and stop what does not pay.

06Common questions about AI ROI.