AI transformation in the Gulf: a field view

The conversation across the Gulf has shifted. Two years ago the question was whether to invest in AI. Now it is how to move a portfolio of pilots into production, and what it takes operationally to keep them there. That is a harder problem and a more interesting one.

Strategy is not the constraint

Saudi Vision 2030, the UAE's national AI programme and Qatar's digital agenda have given organisations mandate and budget. Ambition is not in short supply. The constraint is delivery capacity: teams that can take a promising proof of concept and make it reliable, governed, supportable and integrated with systems that already run the business.

Data residency shapes architecture from day one

In most Gulf engagements, where data physically sits is settled before any model discussion. Regulatory expectations, sovereignty commitments and sector-specific rules mean the architecture question is not simply which model is best, but which models can be run under the required controls — in-region hosting, private deployment, or on-premise for the most sensitive workloads.

This is a strong argument for provider-flexible design. A system that assumes one vendor's public API tends to meet a hard stop the moment it approaches a regulated workload.

Bilingual is a requirement, not a feature

A system that works beautifully in English and awkwardly in Arabic has not been deployed. It has been demonstrated.

• Retrieval quality must be evaluated separately in Arabic — English recall tells you little
• Interfaces need genuine right-to-left support, not a mirrored stylesheet
• Mixed-language documents are the norm, and chunking has to survive them
• Dialect variation matters in anything customer-facing

Local presence still decides procurement

Technical merit gets a supplier onto the shortlist. What decides the award is usually the ability to sit in the room, to support the system in local hours, and to demonstrate accountability through a recognised entity. Buyers here have been burned by remote delivery that evaporated after go-live, and they evaluate accordingly.

The organisations making real progress share a pattern: they picked a small number of operationally boring, high-volume workflows, instrumented them properly, and resisted the pull of the impressive demo. It is a less exciting story than the one in the keynote, and it is the one that reaches production.