← Back to projects

AI / Customer Experience

Real-Time Support Copilot for E-Commerce

Built an AI copilot that drafts support replies using order data, past tickets, and brand voice — reducing response time by 70%.

2024TypeScriptNext.jsOpenAIPineconeShopify API

Context

A Shopify-based e-commerce brand handling 200+ support tickets daily was struggling with response times. Their 4-person support team couldn't keep up during peak seasons.

Problem

Average first-response time was 8 hours. Agents spent most of their time looking up order details and writing repetitive replies. Quality varied wildly between agents.

Approach

Indexed 6 months of resolved tickets into Pinecone for semantic search. Connected Shopify API for real-time order lookups. Built a copilot that drafts replies using context from similar past tickets, order status, and brand voice guidelines.

Build

Next.js interface embedded in the support dashboard. Pinecone vector store for ticket similarity search. Shopify Admin API integration for order context. OpenAI GPT-4 for draft generation with brand voice fine-tuning.

Result

First-response time dropped from 8 hours to 2.5 hours. Agents accepted 80% of AI drafts with minor edits. CSAT scores improved by 15 points within 6 weeks.

What This Proves

AI copilots work best when they augment skilled humans, not replace them. Give agents better tools and the entire customer experience levels up.