E-commerce AI Automation: Where Melbourne Stores Start
E-commerce AI automation in the right order for a small online store: order status first, returns second, what to skip, and what a custom build costs.
Your online store's inbox asks the same five questions every day: where is my order, is this back in stock, can I return it, do you ship to Ballarat, and will it arrive before Saturday. Every answer already sits in Shopify, your shipping app or your returns policy, yet someone is still copying it out by hand at 10pm.
Short answer: The first e-commerce AI automation for most small online stores is order-status and stock replies, because the answers already live in your store and shipping data and the questions arrive in volume. Returns intake comes second, pre-sale product questions third. Skip AI product descriptions and abandoned-cart tinkering until those run, and wait if you get fewer than about ten enquiries a day.
Which e-commerce AI automation comes first, second and third?
Automate in order of volume times certainty: the questions you get most often, whose answers are already in a system, go first. Anything needing taste, discounts or a judgement call about a customer goes last or stays human.
| Order | Automation | Why it is in this position | What it reads from |
|---|---|---|---|
| 1 | "Where is my order" and stock replies | Highest volume, answer is a lookup, wrong answers are rare when the data is clean | Shopify orders, Starshipit, Shippit or Australia Post tracking, stock levels |
| 2 | Returns and exchange intake | Second most common, needs a few rules, and a person approves the outcome | Order history, your returns policy, customer photos |
| 3 | Pre-sale product questions (sizing, compatibility, materials) | Wins sales after hours, but needs good product data first | Product pages, size charts, a short internal notes doc |
| Later | Abandoned cart, review requests, marketplace sync | Useful, but your existing tools already cover most of it | Your email platform and sales channels |
| Skip for now | AI product descriptions on a small catalogue, AI pricing | High quality risk, low time saved for a store under a few hundred SKUs |
The order matters because each step builds on the one before. Once the AI can reliably find an order and read its tracking, returns intake is mostly a few extra rules on top of the same lookup.
Why do order-status replies come before anything clever?
Because they are the questions with a single correct answer that a machine can fetch. A customer who asks "where is order 2041?" wants one sentence and a tracking link. The AI reads the order in Shopify, reads the consignment status from your shipping app, and replies: "Hi Sarah, order 2041 left our warehouse yesterday with Australia Post and is due Thursday. Here is the tracking link."
That reply does two jobs. It answers the question, and it stops the follow-up email that arrives when a customer hears nothing for a day. In most stores the "just checking" follow-ups are a meaningful share of the inbox, and they vanish when the first reply is fast and specific.
Stock questions work the same way, with one rule worth writing down: the AI should only promise a restock date you have entered somewhere it can read. If the date lives in your head, it says "we expect more soon and can email you when it lands" and captures the address.
How should AI handle returns without breaking consumer law?
AI should collect the return details and apply your written policy, while a person approves anything involving money. The policy has to be right before the automation goes near it, because an AI will repeat a bad policy perfectly, thousands of times.
Two points from the ACCC's guidance on repairs, replacements and refunds matter directly to how you script it:
- Change of mind is your policy, not a legal right. Customers are not entitled to a refund under the consumer guarantees just because they changed their mind, but if your store offers a change-of-mind policy, you must follow it.
- "No refunds" wording is off limits. Businesses cannot rely on policies such as "no refunds" or "no refunds on sale items" to deny consumer guarantee rights when a product is faulty.
So the returns flow asks two questions first: is the item faulty or not as described, or is this a change of mind? Faulty items go to a person with the photos and order attached. Change-of-mind requests get checked against your window and condition rules, and the AI drafts the return label request or the polite decline for a person to approve.
What can you do this week without hiring anyone?
You can find out exactly what your inbox is made of in one afternoon, and that alone usually improves your replies.
- Export the last 200 customer messages from Shopify Inbox, Gmail, Instagram and whatever helpdesk you use.
- Tag each one with a single label: order status, stock, returns, product question, or other.
- Write your best answer for the top three labels as saved replies. Include the link a customer actually needs (tracking page, returns form, size chart).
- Fix the root cause of the top label. If "where is my order" dominates, check that your shipping confirmation email includes the tracking link and an expected delivery window. Some of those messages exist because the confirmation email is vague.
That tagging exercise is also the brief for any automation you build later. It tells you whether AI is worth it at all.
Caption: Start with questions that have one right answer sitting in a system you already own.
Is Shopify Inbox or a helpdesk app enough, or do you need custom AI?
An off-the-shelf tool is enough when everything lives in Shopify. A custom AI integration earns its keep when the answer is spread across systems the off-the-shelf tool cannot see.
Off-the-shelf is the right call if your orders, stock and shipping all sit in Shopify and one shipping app. Shopify Inbox and helpdesk tools built for e-commerce, such as Gorgias or Zendesk, connect to Shopify and increasingly include AI features. Check their current pricing and what their AI can read before you build anything.
Custom earns its keep when:
- Stock lives somewhere else. Inventory sits in Cin7, a warehouse system or a 3PL portal, and Shopify's numbers lag behind.
- Your rules are specific. Pre-orders, made-to-order lead times, regional delivery cut-offs, bundles that ship separately, or wholesale customers on different terms.
- Several channels need one brain. Website chat, email and Instagram DMs should all give the same answer from the same data.
- You want the replies in your voice with guardrails you control, such as "never promise a delivery date for anything shipping to WA" or "always hand off orders over $500".
We set out that decision in more depth in chatbot platform vs custom AI agent.
What does a custom e-commerce AI build look like?
A focused first build is an AI agent connected to your store, your shipping app and your inbox, live in two to four weeks. It is not a new store and it does not replace Shopify.
- Week 1: questions and rules. Your tagged messages become the list of question types and approved answers, plus the list of things the AI must never answer alone (refund decisions, complaints, anything legal).
- Week 2: connections. Shopify, the shipping app or carrier tracking, your stock source and the inbox channels.
- Week 3: draft mode. The AI drafts every reply and a person clicks send. This is where you catch edge cases and tune the tone.
- Week 4: auto-send the safe ones. Order-status and stock replies go out on their own. Returns, complaints and high-value orders keep queuing for a person.
Cost, as estimates: most first builds of this kind land between $2,500 and $4,000 at a fixed price, depending on how many systems and channels it connects. Running costs typically sit around $150 to $300 a month, covering hosting (usually under $50) and AI usage that rises with message volume.
Where a person stays in the loop: every refund, every complaint, every message the AI is not confident about, and a weekly skim of auto-sent replies for the first month.
When is AI automation not worth it for an online store yet?
When the inbox is an annoyance rather than a cost. A rule of thumb: under about ten customer enquiries a day, saved replies and a better FAQ page do most of the job for far less. AI also makes things worse when your data is wrong. If stock counts drift or orders get fulfilled outside the system, an AI will confidently tell customers the wrong thing, so clean the data first.
If after-hours sales are the bigger leak, not support load, read why Melbourne retailers lose after-hours sales before deciding where to start.
Frequently asked questions
Will an AI support agent upset customers who want a person?
Not if it hands off cleanly. Customers asking where their parcel is mostly want a fast, correct answer. The ones with a complaint or a damaged item should reach a person quickly with the conversation attached, and a well-built agent recognises those messages and passes them on instead of arguing.
Does a custom AI integration work with Shopify?
Yes. Shopify exposes orders, products, customers and inventory to connected apps, so a custom agent reads the same data your staff see in the admin. The custom part is usually everything around Shopify: the 3PL, the inventory system, your shipping rules and the inboxes.
Can the AI process refunds by itself?
It can, but it should not at first. Let it collect the order, reason and photos and draft the outcome, then have a person approve. Once you trust its judgement on simple change-of-mind returns inside your policy, you can let those through automatically and keep faulty-item claims with a person.
Do we need to stop using our helpdesk to add custom AI?
No. A custom agent can draft replies inside the helpdesk you already use, so your team keeps the same screen. What changes is that the draft is already there, with the order details looked up, when they open the ticket.
Want a second opinion on where your store should start?
We will look at your actual inbox, your store setup and your returns policy, and tell you which single automation is worth building first, including when the answer is "a helpdesk app will do". Book a free AI workflow audit. The audit fee comes off your first project if you go ahead. Rootech is a Melbourne studio and works with stores Australia-wide; see our custom software work in Melbourne.