Why Chatbots Fail with Gulf Customers

Take any Arabic chatbot and send it a sentence a real customer would write — dialect, no punctuation, two questions in one line. You'll find out very quickly which solutions were built for Arabic and which are an English interface with a translation layer.

This isn't a technical footnote. It's the difference between a shop that feels local and one that feels foreign to its own customers.

Where translated solutions collapse

Dialect. A customer doesn't write "what is the price of this product?" in textbook Arabic. They write the Gulf equivalent of "howmuch this?" — compressed, colloquial, and nothing like the phrasing in the training data. A system built on formal-Arabic keyword matching finds nothing and falls back to a generic reply or an apology.

Dialects, plural. The Gulf isn't one dialect. Saudi, Kuwaiti, Emirati, and Levantine speakers use different words for the same question. A model trained mostly on Egyptian Arabic — the most abundant Arabic content online — stumbles on a Saudi customer.

Arabizi. A large share of customers type Arabic in Latin characters with numbers standing in for letters: bkm hatha, wesh 3endkom. Most solutions treat these as meaningless English.

Numerals and dates. Confusion between Arabic-Indic and Western digits, between Hijri and Gregorian calendars, and between date formats. A customer saying "I need it before Eid" requires a system that knows when that is this year.

Mixed-direction text. Arabic and English in one line — "do you have the iPhone 15 Pro in black?" written half in each — produces mangled rendering in a lot of solutions.

Register. Arabic carries sharply different levels of formality. Replying in very formal Arabic to a customer writing in dialect creates distance; replying in heavy slang to a serious enquiry looks unprofessional. Getting that balance right is a linguistic judgement that translation doesn't provide.

A ten-message test

Before buying any solution, send it these, written exactly as your customers write them:

1. "How much is this?" — in dialect, no punctuation
2. "What sizes do you have?"
3. "How many days to Dammam?"
4. "Any discount if I take two?"
5. The same price question in Arabizi
6. "I want to return the order, it didn't suit me"
7. "What time do you open on Friday?"
8. "Do you offer instalments?"
9. "My order is late and I need it urgently" — test whether it catches the frustration
10. "Okay then, how do I pay?"

A good system answers all ten specifically and correctly. A weak one stumbles by the third or fourth, apologises, or — worst of all — invents a confident, wrong answer.

Example. A Riyadh shop deployed a translated global chatbot. The first real message was a dialect phrase meaning "give me your best price on the black set". The system didn't recognise either the verb or the phrase "best price", so it replied with a list of store categories. The customer sent "??" and left. The problem wasn't the AI — it was that the AI wasn't built for this customer.

What matters as much as language

Understanding dialect is necessary but not sufficient. A system that understands the question and doesn't know the answer is no use to you:

Knowledge of your store. It has to read your products, prices, stock, and policies — under their Arabic names, as you wrote them.

Cultural context. Gulf seasons aren't European ones: Ramadan, both Eids, Hajj season, National Day, school holidays. A customer saying "before Eid" is giving you an explicit deadline.

Admitting ignorance. The most important trait of all. A system that says "I don't have that information, let me connect you to the team" is a hundred times more useful than one that invents a price or promises impossible delivery. A fabricated answer doesn't just cost you one sale — it costs you a customer's trust.

Signs of a translated solution

• The settings interface is entirely English, with Arabic offered as a secondary option.
• All examples and documentation are in English.
• It advertises "support for 50+ languages" without ever mentioning dialects.
• It won't give you a real trial in your customers' dialect before purchase.

Always ask for a trial using your own messages, from your real inbox — not a prepared demo.

Summary

Gulf customers don't ask for much: to be understood as they speak, to get a correct answer, and to reach a human when they need one. Solutions that fail here fail because they were built for another market and then translated.

Kasbly was built for the Arabic and Gulf market from the ground up — it understands dialects as they're actually typed, answers from your store data, and admits when it doesn't know instead of inventing.