Skip to content

Illustrative: a retail chain

A WhatsApp agent for order and returns questions, in Arabic dialects

An illustrative scenario: a retail chain’s customers ask about orders, deliveries and returns on WhatsApp, in Saudi and other Arabic dialects and in English, and a small team answers each message by hand. This is how we would build an AI agent that answers from live order data, starts returns within policy, and hands everything else to a person with the conversation attached.

Illustrative scenario

Illustrative scenario: a composite of common requirements, not a description of a specific client project. Figures are targets we would agree and measure against — not results achieved.

The challenge

In this scenario, most customer contact has moved to WhatsApp. The questions repeat — where is my order, when will it arrive, how do I return this — but each one means looking up the order in another system. At peaks such as seasonal sales and Ramadan, replies take hours.

Customers write the way they speak: Saudi dialect, Gulf or Egyptian expressions, English, or a mix in one message, sometimes with a photo. A menu-driven chatbot frustrates them, and a general chatbot that invents a delivery date is worse than no reply.

Our approach

We would start from the conversations themselves: a sample of past WhatsApp chats, grouped by intent, to find which questions are frequent, which can be answered from data, and which must stay with people. The first release covers only the intents that can be answered accurately.

Answers from systems, not from memory

The agent answers order and delivery questions by querying the order management system with the customer’s verified phone number, and returns questions by applying the written returns policy. It can open a return within that policy; refunds, exceptions and complaints go to a person.

A clean handover

When the agent is unsure, the customer asks for a person, or the topic is sensitive, the conversation moves to the service desk with a summary and the full history, so nobody asks the customer to repeat themselves. Outside working hours, the agent tells the customer when a person will reply.

What we would build

What we would build

  • 01

    WhatsApp integration

    A connection through the WhatsApp Business Platform, with approved message templates for order updates and consent recorded for each customer.

  • 02

    Dialect-aware understanding

    A language model tested on real messages in Saudi and Gulf dialects, Modern Standard Arabic, English and mixed text, replying in the customer’s own language.

  • 03

    Tools with limits

    Read access to orders, shipments and store stock; permission to create a return request within policy; no ability to issue refunds or change prices.

  • 04

    Handover to people

    Integration with the service desk — for example Salesforce Service Cloud or ServiceNow — so staff receive the case with a summary, the history and the order already attached.

  • 05

    Quality review

    A weekly sample of conversations checked against what the order system supports. Every wrong answer becomes a test case before the next release.

  • 06

    Privacy controls

    Customer data kept to a minimum in logs, retention agreed with the client, and the agent and its data hosted in the Kingdom; messages themselves still pass through WhatsApp. This supports the client’s obligations under the Personal Data Protection Law but does not by itself deliver compliance, which the client’s legal team confirms.

Target measures

In-scope conversations resolved by the agent
60% or more
Target agreed at kick-off, against current volumes
First reply time
Under 1 minute
Target agreed at kick-off, at any hour
Correct order and returns answers
98% or more
Target, checked on a weekly reviewed sample
Customer satisfaction
At or above the current baseline
Target, measured with a post-chat survey

Expected outcome

The expected outcome is that routine order and returns questions are answered in seconds, at any hour, and the service team spends its time on the conversations that need a person.

These are targets, not results. They are agreed with the customer service lead at kick-off, measured against the current baseline, and reviewed every month.

Have a process that should run itself?

Tell us about it. We will help you work out whether AI is the right answer, what it would take, and what it should return.

Newsletter

Practical AI, in your inbox.

Real use cases, Saudi regulations and what works in Arabic — once or twice a month. No spam; unsubscribe any time.

We’ll e-mail you to confirm. Unsubscribe any time — see our privacy policy.