I’ve been spending more time than usual reading what shop owners say about AI. Reddit threads, Shopify Community posts, odd comments on X. Not as research for a take, really. More because I run an e-commerce business myself, and I’ve been putting agents into the work, so these conversations land differently than they used to.
A few things keep coming up.
The first is tools. Or rather, too many of them. People list subscriptions they barely open, then ask which three are worth keeping. I recognise that feeling. It’s not that the tools are useless. It’s that each one solves a narrow problem, and suddenly you’re paying for five narrow problems instead of one workflow you actually finish. And the thing about subscriptions is that you might not notice the little payments stacking up. Before you know it, you’ve been paying for a tool you stopped using and should have retired half a year ago. (In fact, I bought some lifetime deals for AI tools before 2022, and I hardly touch them any more.)
Then there’s customer service. “Where is my order?” shows up again and again, along with the other routine messages that eat an afternoon. Owners want automation here, but a lot of them push back hard on per-ticket pricing. Fair enough. If your volume is bumpy, a fee that scales with every “where’s my parcel?” starts to feel like a tax on being busy. There’s always a trade-off between time and money (what’s an hour of your day worth, really?), and for a small business it’s often a hard call to make.
The strongest fear, though, is a chatbot saying the wrong thing. Made-up return policies. Refund promises nobody authorised. I’ve seen people say they’d rather answer every ticket themselves than risk that. Don’t get me wrong, I get it. I used to feel the same. Once a bot has told a customer something you can’t stand behind, you’re not debating productivity any more. You’re cleaning up trust.
Copy is another sore spot. AI drafts land in inboxes looking generic, and then someone spends longer editing them into the brand voice than it would have taken to write the thing. I’ve done that myself. The model isn’t useless. It just doesn’t know how you sound until you put a lot of your actual writing in front of it, and even then it drifts.
And then there’s the quiet one: staff pasting customer names, order details, sometimes whole threads into ChatGPT. Privacy protection is real, but I don’t think many people are aware of it. It’s the kind of shortcut that feels harmless until you think about what you’ve just sent out of the building.
Refund abuse and serial returners come up too, though less as a neat tip and more as something people feel in their gut. A few are starting to ask whether AI can spot patterns in reviews or support tickets before a claim turns expensive. That matches what I’ve been noticing on our side: the hard part isn’t generating a reply. It’s knowing which cases need a human, and which ones are following a script.
So what am I left with?
It doesn’t feel like the gap is more prompts. Owners already have prompts. What they want is safer defaults, a clearer sense of where to start, and fewer tools pretending to be a system. Or someone with the experience to set it up so it “just works”, instead of wading through libraries of tutorials, reviews and YouTube videos. I’m still working that out for my own business. The reading helps, mainly because it keeps me honest about where the friction actually is.
