AI Customer Service Agents in Arabic: WhatsApp, Web and Voice

September 20, 2026 · Cortex Solutions

How AI customer service agents in Arabic work on WhatsApp, web chat and voice, what they should handle, and how Saudi businesses can roll them out safely.

Customers now expect an instant answer, in their own language, on the channel they prefer, and in Saudi Arabia that is very often WhatsApp. An AI customer service agent is a system that understands a customer's question in Arabic, written or spoken, looks up your business data, then answers or carries out a defined action, and hands the conversation to a person when needed. This article explains how an agent differs from a classic chatbot, what it should handle, and how to launch one safely under Saudi rules.

Chatbot or AI agent: what is the difference?

A classic chatbot follows a fixed decision tree of buttons and menus; when the customer types something off-script, it stalls. An agent built on a large language model understands many phrasings of the same question, copes with Saudi dialect, spelling mistakes and mixed Arabic-English, and can call real tools: check an order status, book an appointment, or open a support ticket.

The practical difference is that an agent gets things done rather than just replying. That same capability needs guardrails: which actions it may take alone, and which require confirmation from the customer or a staff member.

Where does an AI customer service agent work?

• WhatsApp: the channel closest to Saudi customers. Through the WhatsApp Business Platform (Cloud API) the agent is connected to your official business number.
• Website and app chat: a chat window that knows who the customer is when they are signed in, so it can show their orders and invoices directly.
• Phone and voice: a voice agent converts the caller's speech to text, understands it and replies with natural speech. Useful for clinic and restaurant bookings and for repetitive enquiries outside working hours.

Ideally these channels are several faces of one brain: one knowledge base, one conversation history and shared escalation rules, so a customer never gets two different answers to the same question.

WhatsApp rules every agent must respect

Meta's developer documentation is clear on two points. First, when a customer messages you, a 24-hour customer service window opens and resets with every new message from them; inside it you can reply freely. Second, outside that window you can only start a conversation with a pre-approved message template, and only with people who have opted in to hear from you. Build the agent around these rules from day one, and check Meta's official pricing page, because fees change over time.

What the agent handles and what stays with people

Start with repetitive requests that have a clear answer, and check your logs to see which are most frequent for you:

• Order and delivery status.
• Opening hours, branches, published prices and return policy.
• Booking, changing and cancelling appointments.
• Sending a copy of an invoice or a payment link.
• Collecting complaint details and opening a ticket before passing it to the right team.

People keep sensitive complaints, financial exceptions such as refunds and special discounts, and anything that needs human judgement. The golden rule: the agent knows when to say "I am passing you to a colleague now" and hands over a full summary, so the customer never has to repeat the story.

Practical launch steps

• Pick only three to five use cases for the first phase, and measure their current volume from your conversation logs.
• Prepare the knowledge base: FAQs, policies and product data in clear, up-to-date Arabic. An agent is never more accurate than its sources.
• Connect it to your systems through scoped APIs, your CRM and ERP or point-of-sale system, with read access first and write access only where needed.
• Write the escalation policy: the words and situations that hand the chat straight to a person.
• Test it on real, anonymised conversations from your archive, in the dialects your customers actually use.
• Roll it out gradually, evenings first or a share of conversations, then widen the scope based on results.

Personal data and governance

Customer chats contain personal data: names, mobile numbers, addresses and sometimes health or financial details. Saudi Arabia's Personal Data Protection Law came into force on 14 September 2023, followed by a one-year grace period, and is enforced by the Saudi Data and AI Authority (SDAIA). In practice: collect only what you need, tell customers they are talking to an automated system, decide where conversations are stored and for how long, and review your language-model provider's terms on transferring data outside the Kingdom.

SDAIA has also published AI Ethics Principles, including transparency and explainability, accountability, and privacy. They translate into simple design decisions: a log of every action the agent took, the ability to review its replies, and a named person responsible for its performance.

How to measure success

Do not count how many messages the agent answered; measure outcomes. Track the share of conversations closed without human help and without the customer coming back with the same question, average first-response time, post-chat satisfaction, and the rate of correct hand-offs. Review a sample of conversations every week and update the knowledge base with questions that were answered poorly.

The bottom line

An Arabic AI customer service agent is no longer an experiment, but it is not a product you switch on with one click. Success rests on three things: a narrow scope at the start, sound integration with your systems, and clear governance for data and escalation. At Cortex Solutions we design Arabic chat and voice agents and connect them to our clients' existing ERP and CRM systems. If you are considering this step, we are happy to start by analysing your current conversations and identifying the first use case worth automating.

Sources

• Meta for Developers — Service messages and the customer service window — https://developers.facebook.com/docs/whatsapp/cloud-api/guides/send-messages
• Meta for Developers — Pricing on the WhatsApp Business Platform — https://developers.facebook.com/documentation/business-messaging/whatsapp/pricing
• SDAIA — Personal Data Protection Law — https://sdaia.gov.sa/en/SDAIA/about/Documents/Personal%20Data%20English%20V2-23April2023-%20Reviewed-.pdf
• SDAIA — AI Ethics Principles — https://sdaia.gov.sa/en/SDAIA/about/Documents/ai-principles.pdf