DTC home-goods retailer · AI Support Agents
A direct-to-consumer home-goods retailer was drowning in WISMO and returns tickets every peak season. We built an AI support agent wired into their storefront and 3PL, grounded in their actual policies — and it now resolves seven of every ten tickets end-to-end.
/ 01 — The problem
Support volume tracked revenue almost perfectly: every growth quarter meant another support hire, and every November meant temporary agents answering off-script. First response time averaged six hours in peak weeks, and 'where is my order?' plus returns made up two-thirds of all volume.
The team had tried a rules-based chatbot. Customers learned to type 'agent' immediately — it could talk about orders, but couldn't look at one.
/ 02 — The build
01
We connected the agent to live order, tracking, and customer data before writing a single prompt. Every answer cites what the system actually knows — tracking events, return eligibility, policy text — and the agent refuses to improvise outside it.
02
Return and refund rules were encoded as hard constraints: value limits, day windows, category exclusions. Within the boundary, the agent acts; outside it, a human approves with one click.
03
Two weeks of shadow mode — the agent drafting, humans approving — produced the evaluation data to tune tone and coverage before it spoke unsupervised. Deflection was expanded category by category as quality was proven.
/ 03 — The results
“The bar we set was 'customers shouldn't be able to tell it's not our best agent.' What we got is our best agent — one that answers in nine seconds at 2am on Black Friday.”
/ Free AI audit
Every engagement here started with the free audit. Sennova Labs maps your workflows, prices the waste, and shows you what to automate first.
No obligation · Roadmap is yours to keep
