Agentic AI Is a Top Priority, but the ROI Evidence Is Thinner Than the Headlines

Agentic AI is now the technology a growing number of buyers say they will prioritise. Whether it pays is a separate question, and the evidence available to answer it is thinner than the headlines imply. This article looks at one widely quoted survey, what it does and does not show, and what a business should measure instead of relying on it.

What the Futurum survey found

On 17 February 2026, The Futurum Group published a press release on its 1H 2026 Enterprise Software Decision Maker Survey, based on 830 global IT decision-makers. Its headline claim is that direct financial impact as the primary measure of AI return has nearly doubled to 21.7% of primary responses, made up of 10.6% citing top-line revenue growth and 11.1% citing bottom-line profitability. Productivity as the leading ROI metric fell from 23.8% to 18.0%.

On priorities, autonomous agents were the number one technology priority for 17.1% of respondents, up from 13.0% in the second half of 2025, and a top-two priority for 39.3%, up from 32.0%.

Reading the numbers carefully

The release describes the agentic rise as a 31.5% surge. That is relative growth: 17.1 divided by 13.0. In absolute terms the share naming agents as their top priority rose by about four percentage points. A move from 13.0% to 17.1% is meaningful, and it is also a smaller story than “surge” suggests. For more than four in five respondents, the top priority was still something else.

The ROI figure needs the same care. “Nearly doubled” requires a baseline, and the release does not give the earlier figure. These are also shares of respondents choosing a measure as their primary ROI metric. They describe how buyers say they will judge AI, not how much return AI delivered. A shift from productivity towards revenue and profit is interesting as a change in what executives intend to hold projects to. It is not evidence that revenue and profit rose.

The source matters too. Futurum is a research and advisory firm, and the release promotes a paywalled report. The page does not state fielding dates, respondent countries, company sizes, sampling method or margin of error. Phrases such as “the pilot phase is over” are the analyst’s interpretation rather than survey results. None of this makes the data useless. It makes it directional.

A harder data point from service functions

A more demanding test comes from Gartner. Its survey of 1,303 senior leaders, covered in our piece on the Doorman Fallacy, found service and support leaders committing a median of 12% of their 2025 budget to AI, the highest share of the ten functions assessed, while only 24% demonstrated positive financial returns across their AI use cases. As that article notes, failing to demonstrate a return is not the same as there being none. It does show how many organisations funded the technology before they could show what it did.

Why the evidence is thin

There are structural reasons. Many deployments are recent, so returns have not matured. Costs arrive early and benefits late. Productivity gains are real but diffuse: saved minutes across many staff rarely appear in a ledger unless something is done with them, whether that is fewer hours bought, more work absorbed or faster revenue. And usage-based pricing means the cost side moves with adoption, which can quietly erode the return.

That is also why surveys of what leaders say they measure shift so much from one period to the next. The metric is unsettled because the practice is.

What to measure instead

For a UAE business weighing an agent deployment, I would use four measures, and I would fix them before the pilot rather than after it. Cycle time for the specific workflow, from request to completion. Cost per completed unit of work, including licences, usage charges, integration and review time. Error or rework rate, because an agent that is fast and wrong moves cost around rather than removing it. And a business outcome the workflow feeds, such as conversion, retention or days sales outstanding.

Then compare against a baseline measured on the same workflow before the agent was introduced. If a baseline cannot be measured, the pilot cannot answer the ROI question, and it is better to know that at the start. The same discipline applies on the customer side, as covered in the adoption gap in AI customer service.

A short illustration shows why. Suppose an agent cuts the time to process a supplier invoice from twelve minutes to four. That is a clear productivity gain. But if the same team still handles the same volume with the same headcount, and reviewing the agent’s exceptions now takes longer than it used to, the cost per invoice may be unchanged. The saving only becomes a return when someone decides what to do with the eight minutes.

It is also worth deciding in advance what you will do with a saving. Time recovered that is neither redeployed nor taken out of the cost base is a metric, not a return. A credible business case says which of those it is, who will act on it and by when.

For Dubai businesses this is timely. The emirate’s two-year push on agentic AI, covered in our summary of the announcement, will encourage adoption before the evidence base has caught up. The firms that gain most will be those that can show, with their own numbers, what an agent did for one workflow.

What this does not establish

This article rests on a single vendor-published survey summary. The Futurum release covers global IT decision-makers with no regional breakdown, its figures are self-reported priorities rather than measured returns, and the full report is behind a paywall that I have not read. The Gartner figures are global and cover service and support functions only. Nothing here shows that agentic AI does or does not pay, and no GCC-specific return data is cited. The four measures are my own recommendation, not a finding from either source.

Further reading

Related: The Doorman Fallacy and Enterprise AI covers what a business case leaves out. Why AI Assistant Costs Run Away covers the cost side of the equation. AI Adoption in the UAE sets out the staged framework. How to measure customer experience covers measurement beyond headline scores.

Martin Alva

Marketing, Brand & Digital Transformation Leader, United Arab Emirates

Senior Manager, Employee Experience (Brand Positioning, Digital Platforms & AI) at Space42 (A G42 & Mubadala Company). 15+ years in marketing, brand and digital transformation across MENA and Europe. 5 MENA Effie Awards.

More about Martin Alva → Connect on LinkedIn →