Industries · E-commerce
AI for online stores, built around order questions, returns, peak season and your catalogue
Most AI advice for stores stops at “add a chatbot”. The constraints that decide whether it works are more specific: which platform you run on, how many order-status and returns messages you get, what happens when volume jumps in November, and how many products need accurate data. This page is about those.
Four things that shape every store project
These decide what to build first, how to build it, and when it can safely go live.
Platform reality
Shopify and WooCommerce look alike from the storefront and behave very differently underneath. Shopify exposes one consistent Admin API with published rate limits, so an integration has to queue requests rather than hammer the API mid-sale. WooCommerce runs on your own WordPress hosting, so speed, plugin conflicts and API access depend on your setup, and we check the host before promising response times.
“Where is my order?” and returns
Order-status questions (WISMO) and returns are typically among the largest categories of store support messages. Both can be answered from data you already hold: the order record, its fulfilment status and the carrier’s tracking events. That makes them the safest thing to automate first, because the answer is looked up, not made up.
Peak-season load
Black Friday through the holidays can multiply both orders and messages within days, exactly when your team has the least time. Anything we build is load-tested before peak, falls back to your team if an integration slows down, and is finished well before your busiest weeks rather than launched in the middle of them.
Catalogue size
With hundreds or thousands of SKUs, product data is rarely consistent: missing attributes, thin descriptions, sizes written five different ways. AI is good at extracting and normalising that data and drafting descriptions for review. It also makes support answers and on-site search more accurate, because both depend on the catalogue being right.
Where AI pays off first in a store
In the order we usually recommend, with the data each one needs and the rule that keeps it safe.
Order-status answers
AI chatbots →UsesOrder record, fulfilment status, carrier tracking events.
GuardrailNever promises a delivery date the carrier hasn’t given; delivery problems go to a person.
Returns and exchanges within policy
AI chatbots →UsesOrder date, product-level return rules, your returns app if you use one.
GuardrailExceptions, damaged items and anything outside policy are handed to your team with the order attached.
Catalogue clean-up and descriptions
Workflow automation →UsesSupplier sheets, existing listings, your style guide.
GuardrailDrafts are reviewed before publishing; nothing reaches the live store unapproved.
Supplier and back-office work
Workflow automation →UsesSupplier invoices, purchase orders, stock levels.
GuardrailInvoices are matched and queued for one-click approval, never paid automatically.
Shopify or WooCommerce: what changes in the build
The use cases are the same. The engineering underneath is not.
| Topic | Shopify | WooCommerce |
|---|---|---|
| Where the data lives | In Shopify | In your WordPress database, on your hosting |
| How we connect | A custom app with only the permissions it needs, via the Admin API | The WooCommerce REST API with restricted keys, or a small plugin |
| Performance limits | Published API rate limits: we queue and batch requests | Your hosting: busy periods can slow the store and the API |
| What we check first | Which apps already touch orders (help desk, returns, reviews) | Hosting capacity, caching and plugins that change order data |
Getting ready for peak season
Work back from your busiest weeks. A rushed launch in November is the most common way a store AI project goes wrong.
10–12 weeks before
Choose the first use case, build it and test it with real past messages.
6 weeks before
Load-test at your expected peak volume and agree hand-off rules for the busiest days.
4 weeks before
Live on real traffic and tuned. No new features after this point.
Peak weeks
Change freeze, fallbacks to your team ready, fixes only.
January
Review what peak revealed and choose the next improvement.
Customer data and payments
- Assistants never see or handle card details. Payment questions go to a person or to your payment provider’s own flows.
- Only the order fields needed to answer a question are passed to the AI, and data stays in accounts you control wherever possible.
- Your customers’ data is covered by privacy laws such as UK GDPR for UK shoppers. How personal data is processed, and under which data-processing terms, is confirmed in writing for each project. We don’t claim compliance on your behalf.
What it costs, roughly
An order-status and returns assistant for one store usually fits Starter ($500–$1,500). Multiple stores, several languages, catalogue projects or several connected workflows are typically Growth ($2,000–$8,000). Plan to be live at least four weeks before your peak.
Compare all packages and termsGet ready before your next peak
We reply within 24 hours on weekdays, and after a free call you get a written plan, timeline and fixed quote.