The Problem
A founder-led cosmetics brand had grown from one Instagram shop to selling on Shopify, Instagram Shopping, Daraz, and through WhatsApp DMs. across four channels with completely separate inventory tracking. The founder's mother (who managed operations) was reconciling Shopify orders against Daraz orders against IG DMs every morning to figure out actual stock levels. Errors were costing money: oversold SKUs led to angry customers and refunds; understocking blocked sales the brand could otherwise have made.
Customer service was running entirely through WhatsApp in a mix of Urdu and English, and after-hours messages were going unanswered until morning.
What I Built
A two-layer automation system:
Layer 1. Inventory + Order Sync. A central source-of-truth inventory in PostgreSQL. Every channel (Shopify, Daraz, IG Shopping) pushes orders in via webhooks; the central inventory decrements in real time and pushes updated stock counts back to each channel. Manual reconciliation. eliminated.
Layer 2. Urdu-First Customer Service AI. A Claude-powered agent that handles incoming WhatsApp messages in Urdu, English, or Roman Urdu. The agent has read access to:
- Order status (look up by phone or order number)
- Product info (ingredients, shade matching guidance, application tips)
- Returns/exchange policy
For routine questions ("where is my order?", "is this shade good for medium skin?"), the agent responds instantly. Anything outside its scope is escalated to the founder's mother during business hours.
Outcome
The founder's mother got her mornings back. The brand stopped losing orders to overselling. Off-hours WhatsApp sales. previously dead air. became a meaningful revenue line. The total cost of the automation stack is under PKR 12,000/month including AI usage; the brand recouped it in the first month from oversold-SKU refunds avoided alone.