Arabic chatbots, WhatsApp chatbots and AI voice agents

Customers do not always write the way a script expects. One message can mix Arabic and English, carry a product name in Latin script and ask its question in everyday Arabic. An Arabic chatbot earns its place when it handles that naturally, answers from sources you have approved and knows when to stop. Translating a set of English answers does not make a good chatbot in Arabic.

We start from the conversations that matter to your organisation: the questions your service team answers every day, the status checks that tie up phone lines and the requests that need a record in your CRM. The chatbot answers from your approved content and live data, asks for the details it needs and passes the conversation, with its history, to a person when a question needs judgement.

• Customer service questions answered from your policies, product information and service records
• Order, booking, application and case status read directly from your systems
• Appointments, requests and complaints captured in Dynamics 365 or your CRM
• Internal helpdesks for HR, IT and finance questions, in Arabic and English

A WhatsApp chatbot runs on Meta’s WhatsApp Business Platform, and how you connect to it now matters as much as what the chatbot says. Microsoft has announced that standalone Azure Communication Services capabilities, including its WhatsApp messaging, retire on 30 September 2028. Copilot Studio’s built-in WhatsApp channel relies on that service and stopped accepting new connections after 23 October 2026; agents connected before that date keep working.

For Dynamics 365 Contact Center, Microsoft documents WhatsApp through Twilio as a supported route. Other options include a WhatsApp Business Solution Provider or a direct integration with the WhatsApp Business Platform. We help you choose, and we design the chatbot so that its knowledge and conversation logic sit apart from the channel.

Two policy points also shape the design. Meta’s terms bar AI providers from offering their technology through the platform when AI is the main product, and they restrict using WhatsApp data to train AI models. The ban applies where AI is the primary rather than an incidental or ancillary function, and the terms allow a business to retain an AI provider as its solution provider, so a chatbot that serves your own customers is a different case; we still check each design against the terms in force. Microsoft also notes that publishing an agent to WhatsApp shares agent and chat content with WhatsApp, outside your organisation’s compliance boundary.

An AI voice agent answers calls, understands what the caller says, replies in a natural voice and transfers the call to a person when needed. Microsoft lists Arabic among the languages supported for voice-enabled agents in Copilot Studio, which work with Dynamics 365 Contact Center and support speech and keypad input, context variables and call transfer. Azure Speech in Foundry Tools provides speech recognition and neural voices for Arabic locales including Saudi Arabia, the UAE, Qatar, Jordan, Kuwait and Oman.

Telephony needs the same forward planning as WhatsApp. Microsoft names Teams Phone extensibility as the strategic voice platform for Dynamics 365 Contact Center and lists Teams Calling Plan, Teams Direct Routing and Operator Connect as ways to connect phone numbers. Teams Calling Plans are not sold in Saudi Arabia, the UAE, Qatar, Jordan or Kuwait, so there the options are Teams Direct Routing or, where a participating operator offers it, Operator Connect. For real-time voice agents, Microsoft notes that some speech models are hosted outside the customer’s region and that organisations outside the regions it lists must allow cross-region processing to use them. We review both points with you before design starts.

We have built this pattern into our own product. The AI voice agent in DONOR 360, our charity CRM, discloses that it is virtual, speaks Arabic and English and hands over to a person on request. When we design voice agents for clients, we apply the same principles.

• DONOR 360: charity CRM with an AI voice agent

We take a chatbot or voice agent from its first use case to a live service, then keep improving it. Depending on the task, we build on Copilot Studio or on Azure with Microsoft Foundry and Azure OpenAI, and we use Anthropic’s Claude models where they suit the work.

• Use-case and channel design: which conversations to automate first and where (website, Microsoft Teams, WhatsApp or phone)
• Conversation design in Arabic and English, with terminology, tone and escalation rules agreed with your team
• Answers grounded in your approved content: websites, documents, SharePoint and Dataverse
• Integration with Dynamics 365, Power Platform and your other systems through connectors and APIs
• Hand-over to your staff through Dynamics 365 Customer Service or Contact Center, with the conversation history
• Analytics, content updates and support after go-live through our managed services

• Dynamics 365 CRM and Customer Service
• Managed services and support

Where a chatbot stores and processes data depends on how it is built. Copilot Studio lists the United Arab Emirates (UAE North) among the geographies where it stores customer data. Generative AI is a separate question: for environments hosted in the UAE, Microsoft lists the United States as the location of the Azure OpenAI service used when an administrator allows data movement across regions. Some models are also marked cross-geo, meaning they might process data outside the region, and WhatsApp conversations pass outside your compliance boundary.

For your compliance team, we write down each flow and the setting that controls it, then configure the chatbot for obligations such as Saudi Arabia’s Personal Data Protection Law, the UAE’s Federal Decree-Law No. 45 of 2021 on the protection of personal data and your sector’s rules. Personal data is kept out of prompts unless a request needs it, transcripts are kept only for a period agreed in advance and staff see conversations according to their Microsoft Entra ID roles. Your legal and compliance teams make the final call on how the law applies.

• Azure migration and data residency

Our principle for AI applies to every chatbot we build: AI acts within roles and approvals; people stay accountable. In practice, the chatbot says it is virtual, is configured to answer from the sources you approve and asks a person to approve anything with consequences, such as a refund, a cancellation or a change to a customer record.

Before launch, we test with real questions in both languages, including mixed Arabic–English messages and questions the chatbot should decline. After launch, unanswered questions and hand-overs show where content needs work, and we keep the chatbot current as your products, prices and policies change.

• Our approach: AI with control