What SMEs Actually Ask About AI
On June 25, 2026, I sat on a 44-minute panel at the HKPC E-Commerce Innovation Expo — ninety-plus exhibitors, four panelists from Adaptig, iFlytek, SHOPLINE, and KOLab, moderated by Ken Ip. The prepared talking points were fine. The questions from the SME audience were better, and the panel format didn't leave enough room to answer them properly.
This post is those answers. I sell AI training for a living, so discount what follows accordingly. But these four questions come up at every SME session I've done, and the working answers are more useful written down than compressed into a panel slot.
"I have three staff. Which one learns this?"
Not the youngest person on the team. Not whoever seems "techy." The person closest to the work you'd want to automate — the one doing the repetitive task who would notice if it got easier. They learn alongside their existing job, one tool at a time, starting with whatever eats the most hours. You're not building an AI department. You're giving one person a better way to do something they already do. What falls off the end of their day is the manual version of the task the tool just handled.
"Does this work in Cantonese?"
In mid-2026: mostly, depending on the model, and you have to test it yourself. Here's how. Take a real customer enquiry — the kind that mixes English product names with Chinese descriptions, maybe Cantonese shorthand — and type it into ChatGPT, Claude, or whatever tool you're considering. Read what comes back. Does it sound like something you'd send to a customer, or does it read like a mainland textbook? If the second, try a different model. The code-switching between English product names and Chinese body text that every Hong Kong e-commerce listing relies on is the specific thing most models still handle unevenly. Testing takes ten minutes and saves you from building a workflow on a tool that doesn't match your actual language.
"How do I know it isn't making things up about my products?"
Depends on what you're generating. Marketing copy, social posts, product descriptions in your own words — hallucination risk is low, because the model is working from your input, not citing facts independently. Product specs, prices, dosage, compliance language — the risk is real, and an SME doesn't have a legal team to absorb the fallout. The working rule: anything a customer could dispute or a regulator could question gets a human check before it ships. That's not an AI limitation. It's the same discipline you'd apply to a new hire writing product copy in their first week — except the AI is faster and equally in need of someone reading the output.
"My competitor is using AI already. Am I too late?"
Probably not. Most competitors who claim they "use AI" tried ChatGPT for a social media post six months ago and haven't gone back. The tools in mid-2026 are easier than they were a year ago, and they'll be easier again in six months. Starting now with one real task you do every week — handled partly by a tool — beats having started early and stopped. The window is widening, not closing, because the tools keep getting more practical for small teams that don't have spare time to experiment.
Conference panels — and I say this as someone who was on one — discuss AI in terms of possibility. SME owners ask from constraint: what it costs, how long it takes, whether it works in their language for their products at their scale. The expo listed our session under "innovation." The audience was asking about survival. That's probably the more accurate word for where most small businesses in Hong Kong are with AI right now.
