ServiceNow published its 2026 Enterprise AI Maturity Index on 18 August. It surveyed roughly 4,500 executives and 2,000 employees across 19 countries, compiled by the research firm ThoughtLab, and it contains one number that should frame every AI programme decision taken in the Gulf this year.
Fifty-seven per cent of UAE organisations have implemented agentic AI. Seven per cent have used it to run an autonomous workflow.
The rest of the UAE data is a growth story. AI spending rose 105 per cent year on year. The country's maturity score climbed 13 points to 48 out of 100. The report, covered by The National, calls the UAE one of the world's most ambitious AI markets, and on ambition that is fair. On execution, the same survey is blunt.
The gap is not ambition, and it is not the model
The Index scores organisations across strategy, leadership, data, workflow and governance. UAE firms are strong at the top of that list and thin at the bottom. Seventy-seven per cent of UAE executives told ServiceNow that inadequate data accuracy, access or management is a major obstacle to AI adoption. Fourteen per cent have replaced legacy systems with integrated platforms. Sixteen per cent have put testing, auditing and risk-management processes in place for the AI they are already running.
Read those three figures together and the seven per cent stops being surprising. An agent cannot be trusted to act on its own if the data underneath it is unreliable, the systems it needs to touch are not connected, and nobody can show an auditor how it behaves. So it runs with a person checking its output, which is a copilot, not an autonomous workflow. The pilot succeeds. The business case that justified it does not close.
Saif Mashat, ServiceNow's vice-president for the Middle East and Africa, put it plainly in the report: "The organisations pulling ahead are no longer distinguished by how much they spend on AI, but by how effectively they operationalise it." His shorthand for what does the operationalising: "Strong governance, connected data and orchestrated workflows are what translate investment into business outcomes."
None of that is a modelling problem. It is integration and control work, and it is the part of an AI programme that tends to arrive as a surprise in month five.
What closing the gap actually looks like
The UAE Ministry of Finance reported its first-half 2026 call-centre results in August. After integrating generative AI, sentiment analysis and speech-to-text into the operation, it recorded a first-contact resolution rate of 97.11 per cent against a 90 per cent target, an average speed of answer of eight seconds against a 20-second target, and a customer-happiness score of 95.43 per cent. Call abandonment fell to 1.88 per cent.
Two things about that example matter more than the headline numbers.
First, the AI was folded into an existing service with published targets, not stood up as a separate proof of concept. The generative model answers routine queries around the clock; a document-retrieval tool pulls the right guidance for an agent mid-call; sentiment analysis flags tone. Each piece attaches to a workflow that already had an owner and a metric.
Second, this is assisted resolution, not an autonomous one. There are still agents on the calls. That is the point. The Ministry got a measurable operational result without waiting for the plumbing to reach full autonomy, because it treated the integration and the performance framework as the work rather than the afterthought.
Saudi Arabia has the same problem in a different form
The Index found Saudi employees more resistant to AI adoption than their regional peers, which ServiceNow attributes to more rigid, risk-averse workplaces compounded by siloed data and overlapping tools. The obstacle is again organisational rather than technical. Data that sits in disconnected systems cannot feed an agent that acts across functions, and staff who have seen tools imposed without redesign have a rational reason to distrust the next one.
For firms delivering programmes across the GCC, the read-across is that maturity scores measured at the level of strategy and investment will keep rising faster than the operational reality underneath them. The board sees the ambition number. The delivery team inherits the seven per cent.
What to put in the programme plan
If the objective is AI that is eventually trusted to act, the unglamorous items belong in the plan from the start, not in a later phase:
- Integration into the systems of record the agent must read and write, scoped and costed before build, not discovered during pilot
- An evaluation harness that runs continuously, so drift is visible and the case for widening the agent's autonomy is evidenced rather than argued
- Governance, logging and audit trails in place before scale, because that is what lets a regulator, or your own risk function, sign off on removing the human check
- A single named owner for the decision the AI makes, who reviews it, and how failure is defined and measured
None of this is novel. It is the same discipline that separates AI programmes that ship from AI programmes that demo well. The ServiceNow data simply shows how few organisations in the region have done it yet, and how large the gap is between the money going in and the autonomy coming out.
The bottom line
The UAE's maturity score will almost certainly rise again next year, and the spending figure with it. Neither tells you much. The number worth tracking is the share of deployments an organisation trusts to run without a person in the loop, because that is the one that reflects whether the integration and governance work has actually been done.
Today that number is seven per cent. Closing the distance to something higher is a delivery problem, and it is a solvable one.