Scoping a Kuwait support bot that has to answer in Arabic and English

The job is a support desk that cannot sit on every chat. A customer writes in Arabic or English, the bot answers from your own facts, and a person takes over when the answer is a judgment.

Teams ask for “an AI chatbot for Kuwait.” That is not a brief. Consultancy writes which questions are allowed, which system holds the truth, and who is on the hook when the bot is wrong.

What we walk through first

Where the chats live today — WhatsApp, a site widget, email. Which ten questions actually arrive this week, not the ones on a wishlist. What the bot is allowed to promise (hours, prices, returns) and what it must refuse. How a human sees the thread, in which language, and how they take the keyboard.

If you cannot name the source of truth — a policy PDF, a price list, a CRM note — a fluent model will invent a polite answer. That is a pause.

If the questions are repetitive and the stall is after-hours coverage, a guarded bot can be the first slice. The custom solutions work includes this shape for a Kuwait desk: Arabic and English, instant replies, escalate the rest. We will not treat one company’s FAQ as yours.

This is not the pet-care assistant that has to remember a thread. That job is wellness advice for an animal, with care modes. Company support is a policy and a handoff. We will keep them apart.

When a person should still take the chat

If the customer is angry, if money has to move, or if the answer depends on a record the bot cannot see, a model sits behind a person. If legal has not signed the script, the first slice is a written answer set, not a widget.

If the queue is real and the facts are written down, we scope detect language → retrieve → reply → escalate. That is AI consultancy for the rules, then an agent if you want us to implement.

Bring last week’s ten real chats — Arabic and English — and the page you already send people to. Book thirty minutes or write hello@jamilglobal.com.