12 · AI integration · E-commerce · Automation
Returns & product questions: AI writes the reply
An online retailer answered customer emails about returns and about ordered items by hand — every message with a long thread, many similar cases. That tied up time and expertise in customer service and led to replies worded differently from case to case.
Client project · client anonymized (confidentiality) · project from 2026
- AI integration
- LLM
- Email automation
- Returns
- Product inquiries
- Learning system
- Human-in-the-loop
- Custom development
- > 70%
- time saved handling the emails
- according to the client
- almost 400
- emails per month in the pilot: returns and questions about ordered items
- 4 months
- from idea to a running pilot
01 · Problem
Customers write to the service by email — about returns and exchanges as well as about the items they ordered: features of an article, scope of delivery, usage or the status of an order. Often these are long threads with follow-up questions, photos and changing contacts. Whoever replies has to read the whole thread first, check the order and the rules and then word a suitable answer. That takes time, and depending on who handles it, replies come out differently.
02 · Solution
vona built an AI system that reads and understands the complete email thread — the request, earlier commitments and the status of the order or return — and writes a draft reply from it, whether it is about a return or a question about the item. The team checks the draft, adjusts it where needed and sends it with one click. Every approval and every correction flows back as a learning signal: the system becomes more accurate and its tone moves ever closer to the team’s. Once it is reliably right for a type of request, that level can be switched to autopilot step by step — the AI then replies on its own, while exceptions still go to people. The pilot was up and running after four months.
03 · Result
With almost 400 emails a month, the team saves over 70% of the time according to the client: it no longer reads every thread from scratch but reviews finished drafts. Replies are more consistent and go out faster. With growing experience, the share of requests the system answers without intervention rises — always under the team’s control, with clear rules on when a person decides.
04 · Workflow
- Customer email: Return or question about the order
- AI reads the thread: request, commitments, order status
- Draft reply: written by the AI
- Team reviews: adjusts, sends with one click
- Customer: gets the reply
- EVERY APPROVAL AND CORRECTION BECOMES A LEARNING SIGNAL
- AUTOPILOT, STEP BY STEP
Related services
We can do this for you, too.
The services behind this project — with examples, process and frequently asked questions.
AI integration
We implement artificial intelligence where it creates real value in everyday work — from intelligent chatbots and AI workflows to the automation of recurring office tasks.
Learn more →Development
Web development, app development and custom software — tailor-made, performant and built around your business. We build digital products that keep working: cleanly built, easy to maintain and ready to grow.
Learn more →Keep reading
Background from practice.
Blog
AI in Customer Support: What It Delivers and How to Introduce It
When AI pays off in support, what escalation, knowledge base and measurement look like and how to introduce it in stages without risking quality and data protection.
Read more →Blog
AI Agents in Everyday Work: Uses, Limits and Safety
What sets AI agents apart from chatbots and workflows, when which solution fits, which fields of use have proven themselves and which safety rules you need from the start.
Read more →Next project
Your project here?
Tell us about your problem. The first conversation is of course free of charge and without obligation.
Request a project →