Project Overview:

Built an automation that handles the majority of incoming customer queries on WhatsApp, order tracking, returns/exchanges, and address changes, and offers a seamless “Connect Agent” fallback. It plugs into Shopify for live order data and uses Interakt for compliant messaging.

Outcomes (pilot brand):

• 65–75% automated resolution (deflection) on eligible intents.
• < 60 seconds first response time (from ~10-15 mins earlier).
• +8–12 pts CSAT improvement on WhatsApp conversations.
• 30–40% fewer manual touches for the support team.

Problem Statement:

1. Repetitive queries (WISMO, address change, returns) consumed most agent bandwidth.
2. Response latency varied with shift timing & availability.
3. Fragmented tooling — WhatsApp inbox ≠ Shopify — caused slow lookups & errors.
4. No robust fallback path when customers provided partial or ambiguous details.

Challenges:

• Intent Recognition: Needed a lightweight NLP system to classify customer messages without requiring third-party AI costs.
• Multiple Data Points: Orders could be identified by order ID, email, or phone number, requiring flexible lookup logic.
• Error Handling: Customers often provided incomplete details, so the system had to clarify and re-prompt automatically.
• Dynamic Escalation: Escalation to human agents had to be seamless while preserving the conversation context.