Work & case studies · Retail & E-commerce
ChatGPT-powered support assistant for an online retailer
A GPT-based assistant grounded in order data and help-centre content that now handles the majority of first-line customer inquiries.
Challenge
A fast-growing retailer was answering thousands of order-status, returns and product questions a month with a small support team. Response times stretched past a day at peak, satisfaction was slipping, and the existing rules-based bot answered only exact keyword matches.
What we built
- A retrieval layer over the help centre, policy documents and product catalogue, refreshed nightly.
- A GPT-4o assistant with tool access to the order-management API, so it can look up shipments and start a return, not just describe how to.
- Confidence thresholds and intent-based escalation into Zendesk with a full transcript and suggested reply for the human agent.
- An evaluation set of 400 real questions scored every release, plus a dashboard for resolution rate, hand-off rate and cost per conversation.
Results
Within the first quarter after launch the assistant resolved roughly four in five conversations end-to-end, cut first-response times by more than half, and lifted CSAT on bot-handled conversations by around 40% versus the previous bot. Support headcount stayed flat through a peak season.
What made the difference
Grounding in live order data (not just documents), and treating the human hand-off as a first-class feature rather than a failure state.
The assistant handles the bulk of first-line questions and hands over cleanly when it should. Our team finally has time for the complex cases.
Next case study
Deep-learning market analysis for a fintech
Ready to chart your AI course?
A free 30-minute consultation: we listen, ask hard questions and tell you honestly where AI will and won't help.