Retail
An AI assistant that answers shoppers in Arabic and English
An assistant trained on the retailer’s catalogue and policies, handing complex cases to staff.
- Duration
- 6 months
- Team
- 8 people
- Services
- AI and generative AI, Web
The challenge
What was in the way.
The retailer’s contact centre answered thousands of messages a week, most of them about delivery times, returns and stock. Customers switched freely between Arabic and English, sometimes in the same message. Waiting times climbed sharply during Ramadan and sale periods, and hiring temporary staff each season was slow and expensive. Leadership wanted automation, but not an assistant that would invent policies or promise stock that did not exist.
What we built
The solution.
A bilingual assistant on the website and WhatsApp that answers from the retailer’s own catalogue, order system and approved policy documents. It shows where each answer comes from, checks live order status, and hands the conversation to a person with a summary when it is unsure.
- Web and WhatsApp
- Conversation service
- Retrieval over approved content
- Order and stock APIs
- Agent handover desk
How we worked
The decision that made the difference.
A six week pilot on the top 50 questions, measured weekly, followed by a hardening phase with the retailer’s security team and a staged roll out by channel.
Results
What changed.
- 68%
Queries resolved without an agent
- 4 sec
Average first response time
- 4.6 out of 5
Customer satisfaction score
Most routine questions are now answered instantly in the customer’s language, and staff spend their time on the conversations that need a person.
Technologies used
- Python
- Azure OpenAI
- pgvector
- Next.js
- WhatsApp Business API