AI chatbot for customer service: what it should handle for an Australian small business

Every morning starts the same. Open the laptop, and there's the overnight pile: where's my order, do you ship to Darwin, can I swap these for a size up, do you take Afterpay. You answer them over toast, and by the time you're done the morning's gone.

That's the job an AI chatbot for customer service is built for. Not the tricky stuff, the same five questions, asked a hundred different ways, at hours you're asleep. Get those answered on the spot and your customers are happier and you get your mornings back.

I build AI agents every day, and I build customer service chatbots for Aussie businesses by hand. This is what one should handle, what it should leave to you, what it really costs in Australia, and how to test one before you pay. If you run an online store, the bigger picture is in AI chatbot for ecommerce. This post is the customer service end of it.

What is an AI chatbot for customer service?

An AI chatbot for customer service is a chat window on your website that answers customer questions by itself, at any hour, from your own policies, products and orders.

The old kind was a menu with a chat bubble on it. Press "Shipping", read a paragraph, press "Talk to a human", wait. The new kind reads what the customer actually typed, works out what they need, and writes a proper answer. "Bought boots last week, they're a bit tight, can I swap them and keep the socks?" gets a real reply, not three buttons.

What it answers from is everything. A customer service chatbot that has only skimmed your homepage will guess. One that has your real returns policy, shipping rules, product range and order lookup will get it right.

What can an AI chatbot for customer service handle?

An AI chatbot for customer service can handle most of the everyday questions a small business gets, on the spot, without you. In practice that's:

• Where's my order? Look it up, give the courier and tracking, say when it's due.
• Shipping questions. Do you ship to WA, Tassie, a PO box, a job site? How long, how much?
• Returns and swaps. Your actual rules: how many days, tags on, who pays postage.
• Product questions. Sizes, fit, materials, what's in stock, what goes with what.
• Payments. Afterpay, Zip, PayPal, bank transfer, invoices for businesses.
• Opening hours and pickup. Click and collect, where to park, what to bring.
• Today's deals. The sale, the code in your announcement bar, your price match rules.

For a lot of stores those few questions are most of the inbox. Get them answered in the moment, and what's left for you is the work that actually needs a person.

What should an AI customer service chatbot hand to a human?

An AI customer service chatbot should hand over anything that needs judgement, money or a relationship. A good one knows where that line is and doesn't fight it.

Hand these to a person:

• Complaints. An angry customer wants to be heard by someone who can fix it.
• Refunds outside the policy. "It's 35 days, but I was in hospital" is your call, not the bot's.
• Faulty or damaged goods. Usually needs photos and a decision.
• Custom and bulk orders. Quotes, crew orders, embroidery, trade pricing.
• Anything with a dollar figure you haven't written down.

The handover is where most bots let you down. "Please email support@" is not a handover. A proper one takes the customer's name and contact details, writes you a short summary of what they want, and tells the customer when they'll hear back. You start the morning knowing exactly what each person needs, instead of reading a whole chat to work it out.

Is an AI chatbot or a human better for customer service?

For the repeat questions, an AI chatbot is better: it's instant, it's awake at 2am and it never gets sick of "do you ship to Perth?". For anything personal, a human is better. A small business wants both, each on the job it does best.

Here's the honest bit. If you get five questions a week and you're happy to answer them, you don't need a chatbot. Keep doing it yourself. It earns its keep when the same questions keep landing at hours you can't answer them.

How much does an AI chatbot for customer service cost in Australia?

An AI chatbot for customer service costs anywhere from free to a few hundred dollars a month, and most of them charge per answer, in US dollars.

• Tidio's Lyro: Tidio's own pricing page starts it at US$32.50 a month for 50 AI conversations, rising with volume.
• Gorgias: its pricing page has the Starter helpdesk at US$10 a month for 50 tickets, then about US$1 for each interaction its AI agent handles.
• Intercom's Fin: US$0.99 per resolution on Intercom's pricing page, on top of a paid seat.
• Zendesk: its pricing page lists US$2 per automated resolution on pay as you go, on top of the plan.
• A matey: built and installed free, first month free, then a flat $149 AUD a month. No per-answer billing, cancel with an email.

Per-answer pricing sounds cheap until you do the sums for your busiest month. Black Friday, Christmas and EOFY are when you get the most questions, so that's when the bill is biggest. And because it's in US dollars, the price moves with the exchange rate. Work out what your best month would cost, not your average one.

Why do people hate AI customer service?

People hate AI customer service when it wastes their time: it loops, it guesses, or it won't let them talk to a person. Fair enough. Most of us have been stuck in one.

The fix isn't avoiding AI, it's setting it up properly:

• Answer from your real policies, word for word, so it never invents a rule.
• Say "I don't know" when the answer isn't there, instead of making one up.
• Make the human easy to reach. One ask, and the handover happens.
• Sound like your business, not a call centre.

Getting it wrong costs more than embarrassment. In 2024 a Canadian tribunal made Air Canada pay a customer after its chatbot gave him the wrong refund rule, and the airline's argument that the chatbot was responsible for itself got thrown out. Australian Consumer Law works on the same common sense: what your chatbot tells a customer is your business talking. If it promises a refund your policy doesn't offer, or gets someone's consumer guarantee rights wrong, that's on you.

What does a good AI chatbot for customer service look like?

A good AI chatbot for customer service looks up the order, answers the policy question exactly, and hands the tricky part to a person with the details filled in. Here's a matey on a made-up workwear store, at 7am on a Saturday with nobody in the office:

Look at what it does. It finds the order and gives the tracking without asking for anything else. It quotes the swap rules exactly, including the bit about scuffed soles. And when the boss wants 14 pairs with logos, it doesn't pretend to quote it: it takes the details, raises it with the team and tells the customer when they'll hear back.

How do you set up a customer service chatbot for a small business?

You set up a customer service chatbot by giving it your real policies and products, deciding what it hands over, then installing it on your site with an app or one line of code.

1. Collect your answers. Shipping, returns, warranty, payments, hours. If it's only in your head, write it down now.
2. Connect your products, and your orders if you want it doing "where's my order?".
3. Draw the handover line. List what goes to a person, and where the handover lands: your inbox, your phone, your helpdesk.
4. Install it. An app on Shopify, a plugin on WordPress and WooCommerce, or one line of code anywhere else.
5. Read the chats every week for the first month. Every question it couldn't answer is a gap in your website too.

Step 5 is the one people skip, and it's the one that makes it good. Your range, prices and sales change, so someone has to keep the chatbot up to date. Ask who that is before you buy. With a matey, it's me.

How do you test an AI chatbot for customer service before you buy it?

Test any AI chatbot for customer service with your own customers' questions before you pay for it. It takes ten minutes:

1. Paste in three real questions from your inbox, word for word, typos and all.
2. Ask your trickiest returns question, like sale items or change of mind. Check the answer against your own policy page.
3. Ask where an order is. Does it look it up, or just send you to a tracking page?
4. Ask something your policies don't cover. It should say so, not invent a rule.
5. Get cranky with it. Complain about a late delivery and see if it hands over or keeps arguing.
6. Ask for a human. Count how many messages it takes.
7. Check what you get when it hands over: a summary with contact details, or nothing.
8. Work out the bill for your busiest month.

Or skip the setup. Send me your website and I'll build a demo running on your own homepage, from your real policies and products, before you pay a cent. Then throw all eight tests at it. Want a look first? Try a matey on the homepage.