What shapes an AI project in Lebanon
Language is the first challenge and the main opportunity. A system that copes with real Lebanese messages can take real work off a small team.
| Topic | What it means for your agent | Decide early |
|---|---|---|
| How people write | Lebanese customers type dialect, not formal Arabic, and frequently in Latin letters with numbers standing for Arabic sounds. French and English words appear in the same line | Real conversations to test with, covering every style you receive |
| Messaging as the shop counter | Orders, bookings and complaints arrive by messaging app more than by form or email. The volume lands on a few people's phones | Which channel to start with, and the account it runs on |
| Prices, rates and stock that move | An answer about price or availability is only right if it comes from current data. A stale answer costs a sale or causes a dispute | The source of truth for prices and stock, and how the agent reads it |
| Small teams | The person who answers customers may also pack orders and keep the books. Review work has to fit into their day | Who checks the agent's drafts, and when |
| Customers abroad | Messages come from other time zones at night. An agent can acknowledge and collect details until someone is at work | What the agent may do out of hours, and what waits |
| Cost in foreign currency | Model providers charge by usage in dollars. The saving has to be weighed against that cost, per conversation | A cost ceiling for the pilot, tracked weekly |
| Personal data | Customer names, numbers and addresses pass through the agent. Lebanon's personal data law applies, and a provider abroad will process what is sent | Which details are needed for the task, and which are withheld |
Workflows we build for Lebanese companies
All four begin in a chat window, where Lebanese customers already are.
Order taken from a chat
For restaurants, shops, pharmacies and distributors.
- Customer writes what they want, in their own style
- The agent identifies items and quantities
- It asks one question if something is unclear
- It prepares the order with the total
- A staff member confirms
- The customer receives the confirmation and delivery time
Routine questions answered
For any business that repeats the same replies all day.
- Customer asks about hours, location, price or delivery
- The agent answers from current information
- It replies in the customer's language and script
- It passes the chat to a person when asked or when unsure
Booking by message
For clinics, salons, garages and tutors.
- Customer asks for an appointment
- The agent offers free times from the calendar
- The customer chooses
- The booking is saved and a reminder scheduled
- Changes are handled the same way
After-hours intake
For firms with customers in other time zones.
- A message arrives at night
- The agent acknowledges and gathers the details needed
- It says when a person will reply
- Staff find a summary waiting in the morning
What we build
Agents for messaging channels
Assistants that work inside the channels your customers use, through the channel's official business interface.
Understanding of mixed language
Handling of dialect, Latin-letter Arabic, French and English, tested and tuned on your own conversations. Our guide to Arabic and English AI assistant requirements covers the language questions.
Live prices and stock
A connection to the system or sheet that holds current prices and availability, so answers match reality.
Staff confirmation
A simple review screen or chat command with which staff approve, edit or reject what the agent prepared.
Handover to people
Clear rules for when a person takes over, with the conversation and a summary passed along.
Cost and quality tracking
A weekly view of conversations handled, corrections made and provider cost.
What we need from you
- The channel where most customer messages arrive
- A sample of real conversations, with personal details removed
- Where current prices, stock or free times are kept
- Who will confirm the agent's drafts day to day
- A monthly amount you are prepared to spend on usage
Ownership and licensing terms are defined in the project agreement.
A readiness checklist for an AI pilot in Lebanon
Before the pilot starts
0 of 8 checkedWhat to do next
Tell us where customer messages arrive, what they usually ask and who answers them today. We will reply with a pilot for one task and an estimate of what usage would cost. Start a conversation or book a 30-minute call.
Unsure which kind of system fits? Our comparison of an AI chatbot and an AI agent explains the difference in practical terms.
Common questions
Can AI understand Arabic typed in Latin letters?
Current models handle it better than most people expect, though not perfectly. Spelling is not standardized, numbers stand in for letters, and French or English words are mixed in. Accuracy on your customers' messages is what counts, so we test with a sample of your real conversations before building, and design the agent to ask a short question when it is not sure.
Will the agent confirm orders without anyone checking?
Not at the start. It prepares each order and a staff member confirms with one tap, which catches mistakes while trust is built. Once a category of order has been handled correctly for a period, you may choose to let those pass automatically. Anything involving a complaint, a refund or an unusual request always goes to a person.
What does it cost to run?
Running cost depends on the number and length of conversations and on the model chosen, and providers charge in dollars by usage. We estimate it from your message volume during planning, set a ceiling for the pilot and report actual spend weekly. If the cost per conversation exceeds the value of the time saved, the pilot has answered the question cheaply.
Is this realistic for a small business?
For a business that receives many similar messages every day, yes. The work is narrow: one channel, one task, your own prices and stock. A small pilot shows within a short period whether staff save time. For a business with few messages or highly varied requests, a set of saved replies may be the better investment, and we will say so.
