What shapes an AI project in Saudi Arabia
Two things decide whether an agent works in the Kingdom: how well it handles Arabic as Saudis write it, and whether its use of personal data stands up to the law. The rest follows from the systems it has to connect to.
| Topic | What it means for your agent | Decide early |
|---|---|---|
| AI rules | No dedicated AI law was in force when this page was reviewed. The data and AI authority, SDAIA, has published AI ethics principles and guidance on generative AI, and in 2026 it consulted on a draft Responsible AI Policy that sorts systems by risk | Who in your company follows these texts, and how risky your use case would be judged |
| Personal data | The Personal Data Protection Law applies when an agent reads customer or employee records. It asks for a stated purpose, limits what is collected and sets conditions for moving personal data outside the Kingdom, which includes sending it to a model hosted abroad | Where the model runs, and which fields are masked before a request leaves |
| Arabic as customers write it | Customers write in Saudi dialects, switch to English in the middle of a sentence and send voice notes. Models tend to handle formal Arabic better than dialect, so an agent is measured on real messages from your own channels | A test set drawn from your conversations, and who judges the answers |
| Much customer contact runs through WhatsApp. On its business platform a company may reply freely for a limited period after a customer writes, and needs an opt-in and an approved template to send the first message | The number the agent answers on, and the templates you will need | |
| Stores and business systems | Many Saudi stores run on Salla or Zid, and both publish developer interfaces for orders, products and shipments. Tax invoices stay with the system that already issues them under ZATCA's e-invoicing rules | Which systems the agent may read, and which it may change |
| Regulated and public clients | Banks, health providers and government bodies add conditions of their own, such as hosting inside the Kingdom or evidence of the national cybersecurity controls | Whether a sector rule or a client contract narrows your choices |
Agent workflows we build for Saudi businesses
Four starting points that suit how Saudi companies sell and operate. Our AI agent solutions catalogue lists more, by department and by industry.
WhatsApp order support for a Salla or Zid store
For online stores whose customers ask about an order in Arabic at every hour of the day.
- Customer writes on WhatsApp
- Order found by phone number
- Status read from the store and the courier
- Reply written in the customer's dialect
- Return or address change opened
- Staff take over when asked
Arabic enquiry qualification
For clinics, property offices and training centers that receive more enquiries than they can call back.
- Enquiry arrives from an ad or the website
- Agent asks the qualifying questions in Arabic
- Answers saved to the CRM
- Appointment or callback booked
- Sales receive a short summary
Supplier invoice intake
For finance teams that retype tax invoices into the ERP.
- Invoice received as a PDF or a structured file
- Seller, VAT number and totals read
- Compared with the purchase order
- Differences queued for an accountant
- Approved entry posted to the ERP
Bilingual policy assistant for staff
For HR and operations teams that answer the same policy questions in two languages.
- Employee asks in Arabic or English
- Agent searches the documents that person may see
- Answer given with its source
- Unanswered questions logged for the policy owner
- Gaps in the content reviewed monthly
What we build
Arabic conversational agents
Chat and WhatsApp agents that answer in the customer's dialect and language, grounded in your own content. The WhatsApp agent page shows the pattern.
Store and system connections
Links to Salla, Zid, ERP, CRM and courier systems through their published interfaces, with read and write permissions set task by task.
Evaluation in Arabic
A scored test set built from real Saudi messages, run again whenever the prompt, the model or the content changes.
Human handoff
Fixed points where the agent stops and a person continues, with the conversation and the facts collected so far passed along.
Data controls
Masking of identifiers, a chosen processing region and logs of what was sent, so your privacy position can be shown and not only stated.
Automation around the agent
Rules for the predictable steps and a model for the judgment calls. Our comparison of AI agents and workflow automation explains where the line sits.
What we need from you
- One process with a clear start, a clear end and a known volume
- Real conversations or documents, with names and numbers removed
- Access to the store, CRM or ERP the agent will work with
- Your legal adviser's view on personal data and processing abroad
- An Arabic speaker on your side who can judge the agent's answers
Ownership and licensing terms are defined in the project agreement.
A readiness checklist for an AI pilot in Saudi Arabia
Before the pilot starts
0 of 8 checkedNot sure which process to start with? Our guide to identifying business workflows suitable for AI gives a way to score the candidates.
What to do next
Tell us which conversations or documents take most of your team's time, in which language, and which systems hold the answers. We will reply with the workflow we would automate first and the way we would measure it. Start a conversation or book a 30-minute call.
Common questions
Is there an AI law in Saudi Arabia?
Not a dedicated one, as of this page's review. SDAIA has published AI ethics principles and guidelines for generative AI, which set expectations on fairness, privacy, transparency and human oversight without being binding law. A draft Responsible AI Policy with risk levels went to public consultation in 2026. The Personal Data Protection Law is binding and reaches most agents. Check the current position with your adviser before launch.
Can an AI agent understand Saudi dialect?
Current models read and write Gulf Arabic well enough for most service conversations, though less reliably than formal Arabic or English. Spelling varies, customers mix in English words, and voice notes add a transcription step. So we do not assume. We build a test set from your own messages, score the agent against it and fix the weak spots before any customer sees an answer.
Does customer data have to stay in the Kingdom when we use an AI model?
That depends on the data, your sector and your adviser's reading of the law. The Personal Data Protection Law sets conditions for transfers abroad, and some sectors and public clients require hosting inside Saudi Arabia. The practical options are a provider region your adviser accepts, a model hosted on cloud infrastructure in the Kingdom, and masking personal details before any request is sent.
Can an agent work with our Salla or Zid store?
Yes, within what each platform's interface allows. Both publish developer interfaces through which an agent can look up orders, products and shipment status, and both run app marketplaces. What an agent may change, such as cancelling an order or starting a refund, depends on the platform, your plan and the limits you set. We check this against your own store during discovery.
Where to go from here
Related serviceAI AgentsAgents that work inside your systems, with approval where it matters