Xunodus
All outcomes
Head of Sales · Premium retail

WhatsApp sales agent for a premium retailer

A premium retailer's WhatsApp line was its best sales channel and its biggest bottleneck. We put a bilingual agent in front of it — and gave the sales team their evenings back.

Representative engagement. This case combines patterns from comparable engagements. Client names, figures and interfaces are illustrative.

Sector
Premium retail · GCC
Timeline
11 days to production
Scope
WhatsApp agent, stock and payment integration, human hand-off
Stack
Evolution API · LangChain · n8n · PostgreSQL
Illustrative interfaceArchitecture: inbound webhook, dialect-aware intent, live inventory, payment gateway and the human hand-off queue.
01

The situation

Every campaign ended the same way: hundreds of WhatsApp messages an hour, four sales associates answering from their phones, and replies slipping from minutes to hours. Most questions were the same — is it in stock, in which size, how do I pay.

The cost wasn't only overtime. Buyers who were ready to pay at 11 p.m. went quiet by morning, and the team spent its best selling hours copying stock levels into chat windows.

02

Where it broke

  1. 01Stock lived in the ERP; answers lived in people's heads. Every reply meant switching screens.
  2. 02Payment needed a call back or a store visit, so intent cooled while people waited.
  3. 03No queue and no context: when a conversation changed hands, the customer started over.
Illustrative interfaceWhat the customer sees: stock check, reservation and secure checkout inside one WhatsApp thread.
03

What we built

A bilingual agent sits on the existing WhatsApp number. It understands Gulf and Levantine Arabic as well as English, checks live stock through the ERP, reserves the item and sends a secure payment link — in one conversation, in seconds.

Anything sensitive — a complaint, a custom order, a high-value client — goes to a person through a hand-off queue, with the whole conversation and the customer's history attached. The team picks up exactly where the agent stopped.

System map
  1. WhatsApp
    Customer message, any hour
  2. Agent
    Understands intent in Arabic or English
  3. Stock & payments
    Checks availability, sends a payment link
  4. Sales team
    Receives only the conversations that need a person
04

How we rolled it out

  1. Days 1–2
    Audit of 3,000 past conversations to find the twelve questions behind most of the volume.
  2. Days 3–6
    Agent, stock lookup and payment links built and tested against real conversations, in both languages.
  3. Days 7–9
    Shadow mode: the agent drafted replies, the team approved them, and tone and thresholds were tuned.
  4. Days 10–11
    Live on the main number, with hand-off rules signed off by the head of sales.
Illustrative interfaceThe inventory and payment path, with every step logged.
05

What changed

+38%
Campaign conversion
< 2 s
First reply, any hour
−64%
Repetitive support load

Replies now land in under two seconds at any hour. The team handles the conversations that need judgement — VIP clients, custom requests, complaints — and campaign nights no longer need everyone on their phones.

For the team

The team went back to selling instead of typing the same answer two hundred times a night.

Representative outcomes from comparable engagements.

06

What we'd tell another team

Start with the questions your people are tired of answering. Automating the boring 80% is what makes the remaining 20% feel human again.

Inside the system

Facing the same bottleneck?

Tell us how it works today. You'll get a written architecture sketch — useful even if we never work together.

Message on WhatsApp
Next case study
Operations Lead · Freight & fulfilment
Self-healing order pipeline for a freight network
Read the case study