Context
E-commerce support is a conversion problem disguised as a cost problem: every "where's my order?" or "does it run small?" that goes unanswered is an abandoned cart. ShopAssist is a privacy-safe recreation of a real build — the assistant that answers instantly — in the brand's voice, on the page.
The problem
Generic chatbots damage premium brands. The assistant has to know the actual catalog, the actual order, and the actual size-review data — and know when to hand a $150+ refund or an angry customer to a human.
The approach
I built both sides of the product. The customer side: a polished widget inside a real storefront, resolving an order-status and sizing conversation end-to-end. The merchant side: the training panel — knowledge sources synced from Shopify, escalation rules (refund thresholds, sentiment triggers, failed-resolution limits), and a 30-day performance readout.
What the build shows
The widget answers with shipment dates and size guidance sourced from 340 reviews; the admin view reports 2,940 conversations with 81% resolved without a human. The escalation rules make it enterprise-credible, not just a demo toy.
Recreated for confidentiality — a faithful, fully-working mimic of a real production build; client names, branding and data have been fictionalized.