AI in Saudi Arabia: why projects stall between pilot and production

AI in Saudi Arabia has no shortage of ambition or pilots. The hard part is the step after the pilot. In my experience, most projects stall for organisational reasons rather than technical ones: slow decisions, too few engineers who can run AI in production, data that is not ready, and no clear owner once the demo is over.

I spend much of my week on this problem, both inside my own company, where we automate internal processes, and with customers, where we implement AI automation and agentic AI solutions. The pattern is consistent enough that it is worth writing down.

How many AI projects reach production in the GCC?

The numbers back this up. McKinsey's 2025 survey of GCC executives found that 84% of organisations now use AI in at least one function, but only 31% are scaling it or have fully deployed it, and just 11% get at least 5% of their earnings from AI. BCG's 2026 study of around 200 organisations in Saudi Arabia, the UAE and Qatar put 39% in its "AI leader" group and found Saudi Arabia with a higher share of organisations that are stagnating. Globally, Gartner expects more than 40% of agentic AI projects to be cancelled by 2027.

In Saudi Arabia specifically, enterprises point to the same blockers I see in the field: cost (51%), a lack of skilled staff (47%), integration with legacy systems (42%) and privacy and security (41%), according to a 2025 SAP and YouGov survey. Only 16% of GCC chief executives say their AI tools can reach all the data they need (PwC, 2026).

Why do AI pilots stall in Saudi Arabia?

1. Decisions move slower than the technology

A proof of concept can be built in weeks. Getting it approved for production often takes quarters. Enterprise and government buying in the Kingdom runs through layered approvals, committees and procurement cycles that were designed for multi-year infrastructure projects, not for software that improves every month. By the time a pilot is signed off, the model it was built on has often been superseded. Momentum fades, sponsors move on, and the pilot quietly becomes a slide.

2. There are too few people who can run AI in production

Building a demo and operating a production AI system are different skills. Production needs engineers who can integrate with core systems, handle security reviews, monitor quality and costs, and fix things when a model behaves unexpectedly. That talent is scarce everywhere, and in the Kingdom demand from Vision 2030 programmes is growing faster than supply. The national data and AI strategy targets 20,000 data and AI specialists by 2030, and Cisco's readiness research found 39% of Saudi organisations still lack in-house AI talent. Many organisations rely on a vendor to build the pilot and then have nobody internal to take it over.

3. Data is not ready for the use case

AI is only as good as the data and the systems it can reach. Pilots often run on a clean, hand-picked dataset. Production means connecting to legacy systems, inconsistent records and data that may be classified and subject to residency rules. Questions about where data can be processed, and on which cloud regions inside the Kingdom, regularly surface late and stop a project in its tracks.

4. Nobody owns the outcome

Pilots are frequently owned by an innovation team, while the process they improve belongs to operations. When the pilot ends, it is unclear who funds it, who is accountable for its results and who changes the way the team works. Without that ownership, adoption stays low, and low adoption makes the business case look weak.

5. Change management is underestimated

New tools ask people to work differently. Where teams are comfortable with how things are done, and where there is little incentive to change, even a good system is used reluctantly. The organisations that succeed treat adoption as part of the project from day one, with training, incentives and leaders who use the tool themselves.

What does agentic AI change?

Agentic AI moves from answering questions to completing tasks: an agent reads a request, pulls data from the relevant systems, takes the steps a person would take and hands back a finished result or a decision for approval. That raises the stakes on everything above. An agent needs real system access, clear permissions, an audit trail and a human checkpoint for decisions that matter. It also changes the business case: instead of "a smarter search box", the value is measured in hours of work removed from a process, which is much easier to defend in a budget meeting.

In practice I see agentic AI work best in internal processes first, such as document handling, reporting, sales operations and first-line support triage, where the process is well understood, the data is internal and the risk of a wrong step is manageable.

How do you get AI from pilot to production?

1. Start with a process, not a model. Pick one workflow with a clear owner and a measurable cost, such as hours per week or turnaround time.
2. Agree the production path before the pilot. Settle hosting, data classification, security review and budget owner up front, so a successful pilot has somewhere to go.
3. Design for the team you have. Choose managed services and platforms your people can operate, and plan the handover from vendor to internal team as part of the project.
4. Keep a human in the loop. Let people approve the steps that matter. It builds trust and makes adoption far easier.
5. Measure and publish the result. A number that leadership believes, such as hours saved per month, does more for the next project than any demo.