// Straight answer
Short answer
A good AI agent handles repetitive, well-defined jobs: replying first to enquiries in Arabic or English, asking qualifying questions, booking appointments, sending follow-up reminders, filling the CRM and summarising conversations for staff. It should not set prices, make promises, handle complaints or take decisions where money or reputation is at stake. Start with one workflow.
Answered by Tarek Abou Rjeily, AI consultant & builder in Beirut, for businesses in Lebanon and the Gulf.
In businesses across Lebanon and the Gulf, the weak point is rarely the product. It is the gap between an enquiry arriving on WhatsApp, Instagram or the phone and someone following it up properly. An agent is useful where the task is repetitive, the rules are clear and speed matters.
Do not let an agent negotiate or quote prices on its own, promise delivery dates or outcomes, handle complaints, give medical, legal or financial advice, or make refund decisions. Those need a person, and a confident wrong answer in front of a customer costs more than a slow right one.
Tarek Abou Rjeily, an AI consultant & builder in Beirut, builds agents with human-approval steps where money or reputation is at stake, and does not build a chatbot that answers customers entirely on its own with no human handoff. Also check data rules: the UAE and Saudi Arabia each have their own personal-data laws, so ask any vendor where customer messages are stored and who can read them.
Pick the single process where enquiries most often slip, usually how new leads are captured and followed up, and fix only that. Trying to automate the whole business at once is the most common way these projects fail.
A sensible first step is a short diagnostic. In Tarek Abou Rjeily's Diagnostic, you have one call with the owner and he reviews up to 20 real, redacted conversations, then gives a one-page answer in 3 working days: build, buy, or just fix it by hand. No build is included, and he will say so if the honest answer is that you do not need one.
The next rung is the Lead Memory Sprint: one agreed inquiry path, usually WhatsApp, turned into traceable client records with a named owner and a dated next step, in Arabic and English, in 3 weeks from when access and technical fit are confirmed. Real examples include Invisible CRM, where forwarded messages and voice notes become client records for a Beirut brokerage owner. Measure what changes before widening to a second workflow.
A CRM for a Beirut brokerage owner that fills itself. He forwards messages and voice notes (English, French, Arabizi); it saves the client and answers questions in plain words.
Read case study →White-label CRM for Lebanese small business. WhatsApp screenshots → structured contacts via Claude Vision, bilingual, AI-tagged broadcasts.
Read case study →Not realistically. It takes the repetitive first reply, qualifying and record-keeping, so staff spend time on conversations that need judgment. Plan for a person to review exceptions and approve anything involving money.
Yes, modern models handle both, including mixed messages. Test with real messages from your market, since Gulf dialects differ from Lebanese, and make sure it replies in the language the customer used.
Look for the process with the most enquiries and the weakest follow-up, and count how many recent leads got a clear next step. That count is the baseline a diagnostic starts from.
That depends on the build. Ask where messages are stored, which AI provider processes them, whether data is used for training, who can access records, and how you export or delete them, and check the personal-data rules in your country.
Send the question on WhatsApp and I'll tell you honestly whether it needs building, buying, or fixing by hand.
Ask on WhatsApp →