Open the support inbox of almost any online store and you will see the same questions again and again:
- “Where is my order?”
- “How do I return this?”
- “Can I change my delivery address?”
- “Does this come in a larger size?”
- “Is this compatible with…?”
These questions are important to the customer, but repetitive for your team. AI assistants can now answer many of them well, if they are connected to your real data and set up with clear limits. In this guide I explain how to do that in a way that reduces workload without frustrating customers.
Why old chatbots failed
Many store owners tried a chatbot a few years ago and turned it off. The problems were predictable:
- It answered from a fixed decision tree and could not understand real questions
- It had no access to order data, so it could only say “please contact support”
- It trapped customers in loops with no way to reach a human
Modern AI assistants based on large language models solve the understanding problem. But the other two problems, data access and handover, are still yours to solve. That is where most of the real work is.
Step 1: Find out what your customers actually ask
Export the last three to six months of tickets and group them by intent. You will usually find that a handful of categories make up most of the volume. Typical groups are:
- Order status and tracking
- Returns and refunds
- Product questions (size, material, compatibility)
- Order changes (address, cancellation)
- Payment and invoice questions
- Complaints and damaged items
Each group gets a decision: automate fully, automate with confirmation, or always route to a human.
Step 2: Decide what the AI may and may not do
| Ticket type | Recommended approach |
|---|---|
| Order status | Automate fully, using live order and carrier data |
| Return policy questions | Automate fully, from your policy pages |
| Starting a return | Automate with clear rules and confirmation |
| Product questions | Automate, using product data and specs |
| Address change / cancellation | Automate only if the order has not shipped, otherwise hand over |
| Complaints, damaged goods, legal | Always hand over to a human |
Writing this table down is the most valuable hour of the whole project. It protects you from an assistant that promises refunds it should not promise.
Step 3: Connect the right data
An AI assistant is only as good as the information it can reach. For e-commerce that means:
- Order data from Magento, WooCommerce, Shopify or your ERP, looked up securely by order number and email
- Shipping status from your carriers
- Product data: attributes, size charts, compatibility tables
- Policies: shipping, returns, warranty, payment
- Your help center and past answers from your best agents
The technique behind this is usually called retrieval-augmented generation (RAG): the assistant searches your content first, then writes an answer based only on what it found. Combined with “tools” that call your store’s API for live order data, it can give specific, correct answers instead of generic ones.
Step 4: Design a clean handover to humans
This is what separates a good AI assistant from an annoying one. The assistant should hand over when:
- The customer asks for a person
- The customer is clearly upset
- The question is outside its allowed scope
- It is not confident about the answer
When it hands over, it should pass a short summary and the order number to your helpdesk, so the agent does not have to ask the customer to repeat everything.
Step 5: Launch carefully
- Start on one channel (for example website chat) and one language
- Review conversations daily in the first weeks
- Fix wrong answers by improving the source content, not by adding endless rules
- Expand to email and other languages once the quality is stable
How to measure success
Track these numbers before and after launch:
- Resolution rate: conversations resolved without a human
- Handover rate and handover reasons
- Customer satisfaction (a simple thumbs up/down after each chat)
- First response time
- Agent hours spent on repetitive tickets
Do not optimise for resolution rate alone. An assistant that “resolves” by giving up on customers looks good on a dashboard and bad in reviews.
Privacy and compliance
If you sell in the EU, treat this as a GDPR project too. Only look up orders after verifying the customer (for example order number plus email), log what data is processed, choose AI providers with suitable data processing terms, and tell customers they are talking to an AI assistant. The EU AI Act also includes transparency obligations for AI systems that interact with people, so being clear is both good practice and increasingly a legal expectation.
Frequently asked questions
Will an AI assistant replace my support team?
In most stores it replaces the repetitive part of the work, not the team. Agents spend more time on complaints, VIP customers and problems that need judgement.
How long does a setup take?
A focused first version on one channel can often go live in a few weeks. Most of the time goes into data connections, policies and testing, not the AI itself.
Can it work in multiple languages?
Yes. Modern language models handle many languages well, which is a big advantage for stores selling across Europe. Test each language with real questions before launch.
Want AI support that actually knows your orders?
I build AI support assistants for online stores that use your own order, product and policy data, with clear handover to your team. I am also building Conviro, an AI customer support platform designed for exactly this. If you want to discuss your store, see my AI integration service or request a consultation.