AI automation · Online retail AI for ecommerce: automating support, returns, product data and disputes, with your team in control

Typical workflows in ecommerce

These are typical processes, not client projects. In each, the AI reads the shop data and prepares, and a person decides on exceptions.

1. Order status and delivery questions

“Where is my order?” is the largest group of enquiries in most shops, and the answer is already in the shop and the carrier’s tracking.

• Input: the customer’s message by email, chat or messaging, with order number or email address.
• AI prepares: finds the order, reads the status from the shop and the carrier, and writes the reply in your tone and in the customer’s language. Missing information is asked for in the same reply.
• Support checks: parcels that are late or lost and any message with an angry tone or a complaint.
• Output: an answer sent automatically for plain status questions you have released, and drafts for everything else.

2. Returns and refunds

Customers ask how to return an item, whether it is eligible and when they get their money back.

• Input: the return request, the order and your returns rules.
• AI prepares: checks the order against the rules, creates the returns label or instructions and drafts the reply. It proposes the refund and applies your limits: above them, or outside the rules, a person decides.
• Support checks: refunds above your limit, items outside the window and suspicious patterns.
• Output: a returns case with label and reply, and the refund prepared in your payment system for approval.

3. Product data and feeds

A catalogue with thousands of items needs titles, attributes, images and correct identifiers in every channel.

• Input: supplier data, the catalogue and the rejection messages from your marketplace and merchant feeds.
• AI prepares: consistent titles and attributes, descriptions in each language, a list of records with missing identifiers or mandatory fields, and fixes for rejected items.
• Merchandiser checks: the drafts and every product claim, for example materials, sizes and safety information.
• Output: a cleaned catalogue and updated feeds.

4. Payment disputes and chargebacks

A dispute has a deadline, and the evidence is spread across the shop, the carrier and the customer service history.

• Input: the dispute notice, the order, the delivery proof and the correspondence.
• AI prepares: the evidence pack in the format the payment provider asks for, with delivery confirmation, order details, the terms accepted and the relevant messages in order.
• Team member checks: the pack and submits it. The decision to contest is yours.
• Output: a ready evidence pack and a reminder of the deadline.

What stays with your team, and what we do not do

• Refunds and goodwill: above the limits you set, and for any special case, a person decides.
• Reviews: we never create reviews. Replies to genuine reviews are drafted for a person to send.
• Disputes: the decision to contest and the submission stay with your team.
• Fraud decisions: the AI may flag patterns. Blocking a customer or an order is a decision by a person.
• Decisions about people: we do not build systems that score or select individuals. Systems that assess creditworthiness are high-risk under Annex III of the AI Act, see risk categories.

Rules that matter

The table lists the main distance-selling and product rules as an orientation. Which apply depends on where you sell and what you sell.

On top of this, the GDPR applies to customer data: a data processing agreement with every provider (Art. 28), servers in the EU or UK and a clear rule on retention. Art. 4 of the AI Act asks for sufficient AI literacy among staff, and Art. 50 requires chatbots to say that they are AI. This overview is general information, not legal advice. Your trade body and your solicitor can confirm what applies to your shop.

Systems you probably use

Online shops typically run on Shopify, WooCommerce, Adobe Commerce or Shopware, with a support desk such as Zendesk, Gorgias or Freshdesk, a payment provider and carrier tools, and inventory or order platforms such as Linnworks. We connect through the interfaces these systems offer (API, webhooks, export or email) and work alongside them where none exists. The names are for orientation only and do not imply any partnership.

How AI assistants find online shops

Shoppers ask ChatGPT, Gemini or Google’s AI Overviews for “a durable rucksack for commuting under a set budget” or “where to buy a specific kettle with fast delivery”. Answers draw on product pages, product feeds, structured data, review sites and shopping listings.

• Complete product data: identifiers, brand, price, availability, dimensions and materials in your feed and on the page.
• Structured data: Schema.org Product and Offer that match the visible page.
• Useful content: comparison, sizing and care guides that answer the question a shopper would ask an assistant.
• Genuine reviews: collected after delivery, and shown the same way everywhere.

Our article Ecommerce SEO: product data and AI covers the details, and WhatsApp bot and chatbot for business shows how support can run in messaging.

Is it worth automating? Eight questions

1. Is “where is my order?” the largest group of messages in your inbox?
2. Does every return need someone to look up the rules and write the same reply?
3. Are product titles and attributes inconsistent between your shop and marketplaces?
4. Are items rejected in your feeds without anyone noticing quickly?
5. Do disputes get lost because gathering evidence takes too long?
6. Do support staff switch between the shop, the carrier site and the helpdesk all day?
7. Is there a written rule on refund limits and on what may go into AI tools?
8. Can you name one task where a draft that needs checking would already save time?

Three or more yes answers usually point to a workflow worth testing. Question 7 is the prerequisite.