A fast-growing home goods brand on Shopify had six support agents handling about 4,800 tickets a month, and customers were waiting hours for a first reply.
01The problem
First replies took 4.2 hours on average, and more than 12 hours during product launches. Customer satisfaction slid from 4.6 to 3.9 in six months.
About 70% of tickets were predictable questions, but an earlier rule-based chatbot solved only around 8% before customers learned to skip it.
02What was built
An AI support agent was built to answer customer questions using live order details, shipping information and help content. It can start returns, create labels, apply discounts and update addresses once the customer confirms. Questions it cannot resolve go to a person with a draft reply and relevant context.
Work started with a two-week review of 28,400 past tickets, followed by eight weeks of development. The agent was rolled out in stages, starting with 10% of tickets under full human review.
03How it works
- 01Understand the question
The agent reads each ticket and works out what the customer needs, using live order and shipping details.
- 02Answer only when sure
It replies on its own only when it is confident about both the question and the answer.
- 03Take the action
It can start a return, create a label, apply a discount or update an address once the customer confirms.
- 04Hand over with a draft
Anything it is unsure about goes to a person, with a draft reply and the context ready.
04What changed
Over its first 90 days the agent resolved 64% of tickets, about 3,149 a month, with no human involved. AI replies arrive in about 6 minutes and human replies in 1.8 hours, down from 4.2 hours. Customer satisfaction rose from 3.9 to 4.5.
Monthly support costs fell from $29,200 to $11,100, with three agents moved to other work.
“We’d tried a chatbot before and it was embarrassing — customers hated it.”
Head of Operations (unnamed)
What this could look like for your clients
Your clients’ customers ask the same handful of questions every day. A similar agent could answer them in minutes, handle simple requests itself and pass everything else to staff with a reply already drafted.