Problem & existing process
Hundreds of Messages a Day, Answered One by One.
Before the project, staff answered every WhatsApp message by hand. We analysed the shop's past message history to see what customers actually ask.
- A high daily volume of messages across the shop's numbers, mostly in Arabic.
- The most common message was simply “how much?”, sent right after a product photo.
- Most messages were images, because the shop sells from photo albums.
- Some conversations were never answered at all, which is where sales were lost.
Solution & what we built
An AI Assistant on WhatsApp, and a Panel the Staff Work From.
The assistant handles price, size, availability, hours, payment and delivery questions. Anything else goes to a person with the full chat history.
Prices from the system, never from the AI
Prices, discounts and bank details come from the shop's data and settings. The AI writes the reply, but it cannot make up a number.
Product data synced from the shop's inventory
Products and photos sync automatically from the shop's existing inventory system, so answers follow the current catalogue.
Guided sales flow
Size, then matching designs, then the customer's choice, price, delivery address and an order summary, all inside the WhatsApp chat.
Voice notes understood
Customers' voice messages are transcribed so the assistant can reply to them like text.
Human handoff and shadow mode
Clear rules send a chat to an operator, who is alerted on WhatsApp. In shadow mode the AI only drafts replies for staff to approve or edit.
Staff panel in two languages
Inbox, customers and leads, products, pricing by country, FAQs, sales playbook, blocklist and reports, in Arabic and English, built for phones.
Technologies & capabilities
- WhatsApp Business API
- AI assistant (LLM)
- Voice-to-text
- Next.js
- Supabase / PostgreSQL
- Inventory sync
- Gulf Arabic / English
- PWA
How we delivered
Data First, Then a Careful Rollout.
- 01
Analyse real conversations
We studied the shop's message history to learn the questions, the language and the gaps before writing any bot logic.
- 02
Full panel on sample data
Staff could see and test every screen before the live WhatsApp line was touched.
- 03
Connect the shop's own accounts
WhatsApp, database and AI run on accounts owned by the client, not by us.
- 04
Test number, then shadow, then live
The bot first talked only to a test number while real customers went to operators, so it could be checked safely.
Result
Repetitive Questions Answered, Staff Focused on Real Sales.
Every message is captured
All chats land in one inbox with the customer's profile, so none are lost between phones.
Prices come from the system
Prices and discounts always come from the shop's system, not from the AI, so the assistant does not make up a price or offer an unapproved discount.
The team stays in control
Staff decide when the bot replies on its own, when it only drafts, and when a person takes over.
A clear picture of demand
Reports show what customers ask most, which questions go unanswered and how chats move toward an order.
Related services
- AI Agents & WhatsApp AutomationAn Arabic and English AI assistant on WhatsApp that answers, qualifies leads, updates your CRM and follows up.
- Custom Software DevelopmentBusiness software, CRM systems, dashboards and portals built around the way your company actually works.
- AI & Business AutomationAutomate approvals, back-office tasks, documents and data transfer between your systems, with AI where it helps.
Industries we serve with this
FAQ
Frequently asked questions
Let’s Talk
Is Your WhatsApp Busier Than Your Team Can Handle?
Send us a description of your WhatsApp sales today and we will show you what an assistant could take off your team's hands.
