AI automation for e-commerce

Online retail has the clearest case for automation: real volumes, APIs everywhere and work that grows with every order. We build the workflows that keep support, returns, product data and stock under control as you grow, with refund and brand controls a person can see.

Workflows

Six workflows we build for online shops

Support triage

InputTicket, email, chat, marketplace message
ActionClassify intent, urgency and language; detect repeat contacts and high-value customers
OutputPrioritised, tagged queue; complaints escalated immediately

Order status replies

Input“Where is my order?” with order number or email
ActionLook up order and carrier tracking; draft a reply in your tone with the real delivery estimate
OutputReviewed reply; unattended only for intents you release

Returns and refunds preparation

InputReturn request or portal submission
ActionCheck eligibility rules, generate label, log reason, prepare refund proposal
OutputRefund released by a person; return reasons reported weekly

Product data

InputSupplier spreadsheet, new SKU, incomplete listing
ActionNormalise attributes, draft descriptions from your style guide, flag missing images or sizes
OutputListing draft for review; consistent attributes across channels

Review routing

InputNew review on the shop, marketplace or Google
ActionClassify sentiment and topic; draft a response; route product issues to buying
OutputApproved reply; recurring product problems visible

Inventory alerts

InputStock levels and sales velocity
ActionForecast days of cover per SKU; flag reorder points and dead stock; draft purchase list
OutputWeekly reorder brief for the buyer, no automatic orders

Shopify and WooCommerce

Both have mature APIs and webhooks, which is why they are our starting points. Around them: Gorgias, Zendesk, Freshdesk or Help Scout for support; Klaviyo or Mailchimp for email; DHL, DPD, UPS, Royal Mail and Sendcloud for tracking; Xentral, Billbee, Plentymarkets or a spreadsheet for stock and orders; Amazon and eBay marketplaces via their APIs; Trustpilot, Judge.me and Google for reviews; Stripe, PayPal, Klarna and Mollie for payment status. Names indicate technical compatibility, not partnership.

Shops on Magento, Shopware or a custom platform are possible but usually start with discovery or the audit to check the interfaces.

Refund and brand controls

  • No unattended refunds, credits or order changes. The workflow prepares them with the reason and the policy check; a person releases them. Thresholds by amount are yours to set.
  • Approved templates only. Drafts use your tone-of-voice guide and approved phrasing. The assistant cannot invent a discount or a promise.
  • Complaints go to people. Negative sentiment, damaged goods, chargeback words and legal terms are hard escalation triggers.
  • Peak-season caps. Cost caps and rate limits are set so Black Friday volume does not produce a surprise model bill or a flood of drafts nobody can review.

What we measure

  • First response time by intent, including weekends
  • Tickets per 100 orders and share handled from a draft
  • Days from return request to refund release
  • Listings published with complete attributes
  • Stock-outs on top sellers and dead-stock value

Measured before launch and at 30, 60 and 90 days; see how it works.

Which package fits

One Workflow Pilot (from €1,250): order-status triage and drafted replies, the highest-volume intent in most shops. Solo (€2,750): triage across several intents plus returns preparation. Team (€6,900): support, returns, reviews and inventory alerts with escalation rules and a monitoring view. Multi-brand or multi-country shops with ERP integration are Business scope. Prices excl. VAT and usage; see pricing and the support automation page.

Questions from online retailers

Can the assistant answer customers at night and on weekends?

Yes, for intents you have released after reviewing the accuracy sample, typically order status and standard FAQs. Everything else is drafted and waits for the first agent, or, in the chatbot, creates a ticket with the transcript. Customers are told when they are talking to an assistant and can always reach a person.

Will the AI write product descriptions for the whole catalogue?

It can draft them from your attributes and style guide, and it is good at consistency across hundreds of SKUs. Someone should still review them, and duplicate or thin content across channels is a search-visibility risk we design around. We do not publish anything without a review step you control.

What about customer data and consent?

Order and customer data stay in your shop and helpdesk. Only the text and fields a step needs go to the model provider you choose, with region and terms recorded. Marketing automation stays within the consent your customers have given; we do not build unsolicited outreach. See security and governance.

Our helpdesk already has AI features. Why add this?

Use them where they work. What they usually lack is the connection to your order data, carrier tracking, returns rules and stock, and the escalation and refund controls around it. We integrate with the helpdesk’s features rather than replacing them, and we tell you in discovery if the built-in feature is enough.

Next step

Start with one workflow worth fixing

Answer nine short questions and get an honest read on whether your process is a good automation candidate — and which package fits. No sales call required.