AI chatbots for customer service: what three eCommerce brands learned about hesitant shoppers

What customer service chatbots get right (and what they miss)

Most people don’t leave your website because they’ve gone off your product. They leave because they have a question and there’s nobody to ask. Which size? Will it work with what I already have? How does delivery work? Can it go in the dishwasher?

Those questions tend to turn up in the evening or on the weekend. That’s exactly when people do their considered shopping, and exactly when your team has logged off. An unanswered question quietly becomes an abandoned visit. Nobody complains. They just close the tab.

A customer service chatbot gets the obvious part right. It answers those questions straight away, any time of day, so your team isn’t fielding them by phone, email and social media DMs over and over. It’s the eCommerce version of being asked where the bathroom is forty times a shift, and any relief is welcome.

What most set-ups miss is that many of these aren’t support questions at all. They’re buying questions. “Which size?” is someone with their card half out of their wallet. Treat it as a ticket to close and you save your team a bit of time. Treat it as a sales conversation – in your brand’s voice, with the right answer and a nudge towards the right product – and you win the order. When something genuinely needs a person, a well-trained assistant hands the conversation to your team with the details already attached.

Lesson 1: a good chat assistant pays for itself

AN ONLINE FOOD RETAILER

Buying premium food online for the first time involves a fair bit of hesitating, and it tends to happen at the same points. Which product should I choose? How much should I order? How does frozen delivery work, and will it actually arrive frozen? Our client was answering all of this by phone and email, and losing the sale whenever nobody picked up.

We built a chat assistant with its own name and personality – think of the friendly expert behind the deli counter who knows exactly how many people a kilo will feed. We trained it on the brand’s product pages, FAQs, shipping policies, and storage and preparation guides. Then we added a “Not sure? Ask us!” style button to every product page, so mobile shoppers could find help at the exact moment they were wavering.

Before it went live, the client threw every awkward question they could think of at it. Once customers were using it, we reviewed real conversations every week for the first month and kept refining the answers.

That’s about AUD$12,600 in assisted sales over two months, from a Tidio subscription of around AUD$85 a month. Not a bad return on a team member who never asks for a day off.

A quick word on what “assisted” means, because it matters. These are orders from shoppers who chatted with the assistant. It influenced each sale, but it wasn’t necessarily the only reason for it – someone who opens a chat window is often already fairly keen. The number we find most interesting is the order value. In month two, shoppers who chatted spent noticeably more per order, which suggests the assistant gave them the confidence to buy more. At 20 orders a month it’s early data, but it’s heading in the right direction.

Anything the assistant can’t sort out, such as a delivery problem or a refund request, becomes a ticket for the team. Nothing slips through the cracks.

Lesson 2: the chat tells you what’s really stopping the sale

A KIDS’ SHOE BRAND

Buying kids’ shoes online comes down to one question: what size do they actually need? Ask any parent who has bought a pair in March that no longer fit by June. We set up a chat assistant to help parents work it out, and almost every enquiry was about sizing.

Then the chat logs showed us something we hadn’t expected. When the assistant asked for the child’s foot length, most parents either left the chat or admitted they didn’t have it. The assistant was doing its job perfectly. The parents just couldn’t take the next step – which makes sense if you’ve ever tried to get a four-year-old to stand still next to a tape measure.

So we changed the approach. We created a free printable foot measuring chart that parents can use at home in a couple of minutes. We trained the assistant to offer the chart at the exact moment a parent gets stuck, rather than letting the conversation end there. And we taught it to explain why measuring matters. There’s no standardised kids’ shoe size chart in Australia, so a size from one brand isn’t a reliable guide to the same size from another.

Today, around one in four shoppers who chat with the assistant go on to buy. That’s about 25%, compared with a site-wide conversion rate of around 2%. It isn’t a perfect like-for-like comparison, because people who start a chat are already more interested than the average visitor. But it’s a very big gap.

The bigger lesson is that a chat assistant doubles as a research tool. Every conversation is a customer telling you, in their own words, exactly what they’re unsure about. Here, it led to a fix the brand would never have found by staring at a dashboard. With the food retailer, the chats revealed gaps on the product pages – nutritional details, for example, that only appeared inside images, where shoppers skimming on their phones were easily missing them.

Lesson 3: get it working before the busy season, not during it

A CHILDREN’S LUNCHWARE BRAND

This brand’s customers ask very specific questions. Which lunch box fits inside which lunch bag? Is this bottle dishwasher-safe? What’s the difference between a vinyl name sticker and laser engraving? Until recently, every one of those questions landed in the team’s Shopify Inbox.

Black Friday, Christmas and Back to School are the brand’s three biggest trading periods, and they arrive back to back. That’s precisely when repetitive questions pile up and the team has the least time to answer them. Launching a chat assistant in the middle of all that would be like changing a tyre while you’re still driving. So we planned the project around one goal: discovery, build and launch within the first week, so the assistant is tested and settled well before Black Friday.

We’ve trained it on the brand’s existing fit guide, lunch box and lunch bag size charts, care instructions, shipping thresholds, personalisation options and bundles. Weekly reviews will continue right through the peak season, so the assistant keeps up as the questions change – and they will. We’ll share the results in a future update.

How we set up an AI chat assistant that actually sells

Every brand is different, but our process follows the same four steps.

Discovery and knowledge base. We find the questions your customers keep asking, then train the assistant on your product pages, size guides, FAQs, policies and any other resources you’d like it to draw on.

Brand voice and placement. We give the assistant a name and personality that suit your brand, and add prompts on your product pages so mobile shoppers can find it when they need it – not tucked away in a corner nobody looks at.

Testing before launch. You can ask it as many tricky questions as you like before a single customer sees it. We also set clear rules for when it should hand a conversation to your team.

Weekly optimisation and reporting. For the first four weeks, we review real conversations every week, fill the gaps and fine-tune its answers. You get a performance summary and a short training video, so your team can manage it confidently from there.

Why not just do it yourself?

You certainly can start there. Free tools like Shopify Inbox are a reasonable first step, and they install in minutes. But as your chat volume grows, the limits start to show.

Shopify Inbox is built for chat on your website, so your Facebook and Instagram messages live somewhere else. When the AI can’t answer, there’s no structured ticketing process for your team to follow up. You can set a general tone, but not detailed rules about what the assistant should never claim, or when it should bring in a person. And like any new team member, a chat assistant is only as good as its training. Someone has to keep reviewing and improving it.

The details matter too. With the food retailer, for example, we worked out how to send escalated chats to the team as tickets without flooding their main inbox with every customer email. It’s those small decisions that decide whether your team comes to trust the assistant, or quietly switches it off.