From AI Adoption to AI Operations: What Government and Banking Need in Place Before Scaling
Explore why GCC AI adoption is outpacing scaled deployment and what data, integration, ownership, auditability, and KPIs are needed to move from pilot to production.
Reported AI adoption in the Gulf is high, and most of the figures describe usage. In McKinsey’s survey of 139 senior executives and board directors across the six GCC countries, fielded in August and September 2025, 84% said their organisation had adopted AI in at least one business function. Only 31% said AI was being scaled or fully deployed across the organisation. And 11% qualified as “value realisers”: organisations that have adopted AI, are scaling it, and can attribute at least 5% of earnings to it. The survey is self-reported and spans five sector groups, including financial services and the public and social sectors. It shows the direction of the market; it does not describe any single ministry or bank.
For a CIO or CDTO in government or banking, the constraint sits between the first number and the second. One kind of use case runs in a pilot. Another has an owner, data access, an audit trail and a target in daily operation. The sections below separate what the adoption figures measure, show how the gap appears in Qatar, Saudi Arabia and the UAE, and list what has to be in operation before a use case is scaled.
What the adoption figures measure
Three kinds of figures circulate in the region, and each answers a different question.
Usage telemetry. Microsoft’s AI Diffusion Report estimates the share of people who used a generative AI product, from aggregated Microsoft telemetry adjusted for device share, internet penetration and population. For the second half of 2025 it puts the UAE at 64.0% and Qatar at 38.3%, up from 59.4% and 35.7% in the first half (Microsoft, January 2026). Microsoft notes that no single metric is perfect. The metric counts people using tools. It says nothing about whether an organisation has placed those tools inside a process.
Employee surveys. BCG’s AI at Work survey (more than 1,000 participants in Kuwait, Qatar, Saudi Arabia and the UAE, published 25 September 2025) found that 86% of respondents report AI agents already in pilots or integrated into workflows, against 69% globally. The figure groups pilots and integrated workflows together. The same respondents named decisions taken without human oversight and unclear accountability for mistakes among the risks.
Executive surveys. Deloitte and MBZUAI (February 2025; more than 150 business and technology leaders in the UAE, Saudi Arabia and Qatar) reported that more than 80% of organisations feel pressure to adopt AI, while almost half say they lack the talent and technology capabilities needed to scale it.
The evidence points the same way in each case: usage is wide, and scaled, measured operation is a smaller group. McKinsey’s 31% and 11% describe that group for the Gulf as a whole. A broader maturity scorecard is in GCC AI Maturity Benchmark 2026.
Three markets, three starting points
Qatar. In September 2026, an expert in strategy and government development told Qatar Television that about 1,650 of roughly 2,300 government services are fully digitised and integrated (The Peninsula, 15 September 2026). The figure is his own estimate. He said the number of online services should not be treated as the ultimate measure of success. He gave a hypothetical example: if a citizen still has to enter data on five platforms and wait for an employee, then, in his words, the government has “digitised the problem rather than solved it”. He proposed measuring time saved, forms and procedures removed, and whether users complete a service at the first attempt. Among the next challenges he listed data silos and differences between government systems and standards.
At a cloud and AI summit in Doha on 22 September 2026, an official of the Ministry of Communications and Information Technology said cloud infrastructure provides the foundation to integrate advanced and agentic AI capabilities into core government operations. The Council of Ministers highlighted the deployment of a centralised data analytics infrastructure and an Enterprise Data Lake for executive reporting (Qatar News Agency). For a Qatari government CIO, the practical requirement is to know, for every AI-supported service, which entities’ data it draws on, who owns the outcome and which measure shows that the service works. The wider government picture across the GCC is covered in Five AI Trends Shaping Government in the GCC.
Saudi Arabia. The Digital Government Authority runs an annual Digital Transformation Measurement (Qiyas). The 2025 cycle assessed 240 government agencies against 95 criteria covering, among other areas, integration and operations, and the overall index score was 88.30%, up from 87.14% in 2024 (DGA). The index measures how far agencies have advanced in digital transformation. It does not count AI use cases in production, so it cannot answer the question this article asks. For AI use cases, in the author’s view, the KPI should be defined before deployment and reported on the same cycle as the rest of the digital programme, so that results can be compared and checked.
United Arab Emirates. At national level, the Microsoft figure of 64.0% (second half of 2025) means that individual use of AI tools is already widespread. BCG’s survey covers the four GCC countries it included, and across them 63% of respondents would use AI tools without company authorisation. That number covers the region and says nothing UAE-specific. Where individual use is this common, the operating task is to identify which uses sit inside a process that has an owner and a record, and which do not. The sources cited here are national or GCC-wide. Abu Dhabi and Dubai have separate public-sector contexts, and this article keeps them apart.
What has to be in operation before an AI use case scales
McKinsey’s GCC report found that the organisations it classed as value realisers rated themselves stronger on operating model, technology and data, and change management. That subgroup is small: 13 respondents answered the question. In interviews with 14 senior executives, the same report described performance checks that tie deployment to measurable KPIs such as cost savings, revenue uplift or cycle-time reduction. In the author’s view, the five items below are the test to apply. The sources cited do not set them out. For a use case in a ministry, an authority or a bank, each item can be checked before the scaling decision. Each needs evidence; a statement of intent does not count. Failure patterns between pilot and production are described in Why 70% of AI Projects Fail and From Pilot to Profit.
- Data. A named list of the sources the use case reads, an owner for each source, documented access rights and a data quality check that runs on a schedule. Test: can the team list every source, with its owner, on one page?
- Integration. The output reaches the system where the work happens, such as the case management, core banking or ticketing system, through a maintained interface. Exported files and manual re-entry indicate that the use case is still a pilot.
- Process owner. One named business owner is accountable for the outcome and has the authority to pause or stop the use case. IT owns the platform, and the business owns the result.
- Decision log. A record of what the system recommended or did, on which data, and who approved, changed or overrode it. BCG’s respondents named decisions without human oversight and unclear accountability for mistakes among the risks; the log is the working answer to both.
- Measurable KPI. A baseline measured before deployment, a target, a reporting cycle and a stop condition. In government, the measures proposed by the Qatari expert (time saved, forms removed, first-attempt completion) are examples of outcome KPIs. In banking, examples are cycle time, handling cost and error rate for the specific process, each compared with the pre-deployment baseline.
In the author’s view, a use case that lacks any one of the five is still a pilot, whatever its user numbers. For banks, items 3 and 4 matter most where a model output affects a customer: a named owner and a record of the decision are the minimum. Sector context is in The State of AI in GCC Financial Services.
Where integration carries the work
Items 1 and 2 depend on how the underlying systems are connected. Systems of record in a ministry or bank are often built at different times, by different vendors, with different data definitions, and an AI use case typically needs consistent data from several of them at once. Usetech’s Usebus AI-Code combines ESB, ETL and orchestration, on-premises or in the cloud, and turns fragmented systems into an operationally connected environment where data moves fast enough to support real decisions and automation. The role of the integration layer in AI platforms is described in Why AI Platforms Need a Reliable Data Integration Layer.
The test for the next steering committee
For each AI use case reported as live, ask for the five items. Use cases that can produce them are candidates for scaling. Use cases that cannot should be reported as pilots, with a date by which the missing items will be in place. This gives a CIO or CDTO a status that can be defended in front of a steering committee or a board: the adoption figure stays accurate, and the scaled count rests on evidence.
