// case study
Support replies in 30 seconds instead of 6 hours
An AI assistant trained on product data and order status now answers most customer questions instantly and hands the rest to a human with full context.
- The client
- Industry
- Retail & e-commerce
- Services
- AI Solutions, Web and Digital Products
- Year
The client
An outdoor-gear retailer with 22 employees, a busy web shop and a three-person customer service team.
Their X
“Every morning we open the inbox to 200 emails, and most of them are the same ten questions about sizes, delivery and returns.”
How we understood it
We read 1,200 real tickets and found that 64% were answerable from three sources: the product catalogue, the order system and the returns policy.
The remaining questions needed judgement (damaged goods, complaints, wholesale), and those were exactly the ones the team wanted more time for.
So the goal was not “a chatbot”. It was: answer the routine questions instantly, and give humans the hard ones with everything they need already collected.
What we built
A chat assistant on the website and in the email flow that answers from the live product catalogue and order status, in Swedish and English.
When a question needs a person, the assistant writes a summary, attaches the order and hands it to the helpdesk queue.
Key decisions
- Answers only from approved sources; says “I don't know” instead of guessing.
- Order lookups require the order number and email, so no personal data leaks.
- All conversation data stays in the EU.
- Web shop chat
- Support inbox
- AI assistant
- Catalogue + policies
- Order system
- Helpdesk (humans)
- Web shop chat → AI assistant
- Support inbox → AI assistant
- AI assistant → Catalogue + policies
- AI assistant → Order system
- AI assistant → Helpdesk (humans)
Results
- Median first reply
- 0.5 mins
- Before: 6 hrsAfter: 0.5 mins
- Questions answered without a human
- 61%
- Before: 0%After: 61%
- Support emails per day
- 78
- Before: 200After: 78
In their words
“They spent the first week reading our tickets instead of pitching tools. The assistant now handles most of the routine questions, and our team finally has time for the customers who need us.”
Built with
- Claude
- Next.js
- PostgreSQL
- Zendesk
- Shopify
// next step
Have a similar X? Let's talk.
Tell us the problem in your own words. We'll reply within one business day with a straight answer on whether we can solve it, and how.
Free · 30 minutes · no commitment