The short answer
- A modern AI assistant understands and replies in the customer's own dialect, and switches to English when the customer does.
- It answers from your material, such as price lists, policies and FAQs, not from the open internet, so it does not invent prices.
- It should capture the lead, book the appointment and hand over to a person when the conversation needs one, from the same inbox your team uses.
- Your data should be isolated, exportable and never used to train models.
Why translated chatbots fail in Arabic
A customer in Cairo writes "عايز أحجز بكرة الصبح"; a customer in Riyadh writes "أبغى موعد بكرة الصباح". Older keyword bots need both sentences to match a script, so they fail at the first spelling they did not expect. Customers notice within one message and ask for a person.
An assistant built on a large language model reads the sentence the way a native speaker does. It understands Egyptian, Gulf, Levantine and Maghrebi dialects, mixed Arabic and English, and the shorthand people actually type, and it replies in the same register.
How it learns your business
The assistant does not browse the internet for answers. During implementation your price list, policies, working hours, product list and frequently asked questions are added to a knowledge base as text, PDF, Word or text files. Product feeds can be synced from a URL or an API on a schedule so prices stay current.
Instructions set the tone and the rules: how formal to be, what it may promise, when to ask for a handover. When a question falls outside the material, a well-configured assistant says so and offers a person, instead of guessing.
What a good assistant does beyond answering
- Captures the lead: name, phone, interest and status, without a form.
- Books appointments inside your working hours and gives a reference.
- Follows up with people who asked and went quiet, a limited number of times.
- Reads voice notes, photos and PDFs the customer sends.
- Replies with buttons, cards and lists where the channel supports them.
When should it hand over to a person?
Honest automation means the assistant never pretends to be human and knows its limits. It should request a handover when the customer asks for one, when the question needs judgement, such as a complaint or a medical or legal question, and when a sale is big enough to deserve a human conversation.
The handover should be seamless: the assistant pauses, your colleague answers from the same inbox, and hands the conversation back when done. A mood score that flags unhappy customers helps your team pick the right conversations first.
Questions to ask about your data
Ask where conversations are stored, whether they are isolated from other businesses, whether you can export leads and contacts at any time, and whether anything is used to train AI models. The answers should be: isolated per workspace, exportable, and never used for training.
If your policy requires data to stay inside your infrastructure, ask whether the platform can be installed on your own servers, and what still has to be public. For WhatsApp, Meta needs a public HTTPS endpoint to deliver messages in any setup.
How aiyourbusiness does it
aiyourbusiness is Arabic first: the dashboard, the widget and the assistant work right-to-left by default with English available. The assistant replies in the customer's dialect, answers from your knowledge base, captures leads, books in Google Calendar and hands over from a unified inbox. Data is isolated per workspace, exportable, and never used to train models, and the platform can run in the cloud or on your servers.