From automation to advantage

AI is not yet fluent enough

The region has made significant progress in consolidating finance, HR, IT, procurement, and customer operations into centralized business services. Expectations now extend beyond efficiency to quality at scale. AI is the key, yet it remains limited by language and context.

Large language models are trained predominantly on English data. In major datasets such as Common Crawl’s CC100, , which is used to train mainstream LLMs, english accounts for roughly 82GB of text compared to just 5.4GB in Arabic, despite more than 400 million Arabic speakers.¹ Dialect diversity, code-switching, and linguistic complexity widen the gap between capability and performance.

The impact is evident across the GCC. Pilots demonstrate potential but struggle to scale. Chatbots require frequent human intervention, document-processing tools misinterpret text, and automation often merely shifts effort without eliminating it. Generic Arabic-capable models achieve only around 30-50% accuracy on Arabic financial text.²

This is not due to AI capability, but a limitation of enterprise-grade Arabic fluency. More mature ecosystems such as China have reduced this gap through investment in native-language models, data and deployment.

How Arabic-first AI changes the economics

When models are not natively fluent, organizations compensate through supervision, rework, and parallel manual processes, eroding the advantage of automation. More than 20% of Arabic chatbot queries still require human handover.³

Arabic-native LLMs are still nascent, and mostly limited to front-end conversational layers. AI has struggled to move into end-to-end workflow automation, where impact compounds across processes.

Improved language alignment can facilitate a shift that is already underway. IBM’s watsonx Orchestrate, for example, automates finance journal entries to generate approximately $600,000 in annual savings, reduces HR costs by around 40 percent, and enables a tenfold acceleration of supplier onboarding. AI operates here as an integrated execution layer across workflows.

This transition is evident in China, where AI is embedded across enterprise and public-sector processes and improves through real-world feedback loops.

The GCC is beginning this journey; initiatives such as ALLaM, JAIS, Falcon, and Fanar, alongside developments like the OpenAI-G42 collaboration on a customized government model, reflect a shift toward Arabic-native, institution-specific AI. Adoption, however, remains concentrated at the conversational layer.

Talent is the real multiplier

Technology does not scale on its own. The effectiveness of AI depends as much on talent as on model capability.

The region has deep Arabic-speaking talent pools. When concentrated in transactional roles, the impact of this talent remains limited. But when positioned at control points – AI governance, data stewardship, model supervision, and service design – talent improves performance and enables scale.

China offers a clear example of what this looks like. With talent demand projected to reach around six million professionals against around one million available, it has increased investment in education and workforce transformation.⁴ Specialist institutions undertake research, while universities and platforms led by firms such as Baidu, Alibaba, and Tencent handle the deployment. The GCC is building towards this. Institutions such as MBZUAI and SDAIA are cultivating talent and there is now a growing opportunity to channel that pipeline to enterprise deployment.

The most resilient model is hybrid by design: regional talent provides scale, while national professionals anchor governance and oversight. Indeed, advantage will depend not on access to AI, but on the ability to localize, govern, and operate it effectively.

From automation to advantage

The GCC does not need to replicate India’s outsourcing model or Europe’s nearshore approach. Its advantage lies in strong state capacity, rapid infrastructure build-out, and the ability to align policy, capital, and demand.

Converting this into sustained advantage requires two structural shifts.

First, Arabic-first AI must be treated as core infrastructure – spanning data, models, compute, governance, and integration into workflows. Without this, AI remains fragmented.

Second, workforce strategy must move from participation to control, positioning national talent in roles that define how systems operate rather than being limited to operating processes built elsewhere.