Case study · E-commerce

A support chatbot that returns 14% of revenue.

A bilingual shopping and support assistant over an 18,000-SKU catalogue — replacing three contact-centre tools with one conversation.

Client
Oak & Roam
Sector
Outdoor retail
Built in
5 weeks
Live since
Nov 2025
AOV recovered
+14%
AOV recovered
from in-chat upsell
tickets resolved
71%
tickets resolved
with no human handoff
languages
2
languages
English + Spanish
median reply
<2s
median reply
streaming answers
The problem

Support was a cost centre that lost sales.

Customers asked sizing and returns questions across email, chat and social. Average first reply was 9 hours, and a third of carts with an open question were abandoned before anyone answered.

The approach

RAG over the whole catalogue.

We indexed products, policies and 40k past tickets, then built a brand-voiced agent that can look up orders, start returns and recommend alternatives — with a clean handoff to staff when it is unsure.

  1. Ingestion
    catalogue · policies · ticket history
  2. Retrieval
    pgvector · hybrid search
  3. Agent
    orders · returns · recommendations
  4. Handoff
    Zendesk with full context
The outcome

Faster answers, bigger baskets.

Seven in ten conversations now resolve without a human. Because the assistant recommends in-stock alternatives instead of saying "out of stock", average order value on assisted carts is up 14%.

It sounds like us, it knows every product, and it never sleeps. Our team now spends its time on the conversations that matter.

Aiyana Patel · Founder, Oak & Roam