Should Your Small Business Use an AI Chatbot? An Honest Assessment
The pitch lands in every small business owner’s inbox eventually: an AI chatbot that answers your customers 24/7, never takes a sick day, and costs less than a part-time hire. Some of that pitch is true. Enough of it is oversold that we think most small businesses deserve a more honest assessment before they buy.
We help Vancouver businesses with their digital presence, and we’re not anti-chatbot — modern AI chat agents are dramatically better than the decision-tree widgets of five years ago. But “better” isn’t the same as “right for you.” Here’s how we’d actually decide.
What AI chatbots genuinely do well
Give a modern chat agent your real business information and it handles the routine layer of customer contact competently:
- Answering repetitive questions instantly, at any hour. Hours, location, parking, pricing ranges, booking links, what to bring to an appointment. In our experience these questions make up a large share of most small businesses’ enquiries, and none of them need a human.
- Capturing after-hours leads. Someone comparing three restaurants or clinics at 10pm won’t wait for your 9am reply. A bot that answers the basics and takes a name and number turns a lost visitor into a morning follow-up. One catch: a chat enquiry is not a newsletter sign-up. Under Canada’s anti-spam law an enquiry gives you implied consent for only six months (CASL, s. 10), so ask before adding anyone to your list, as our email marketing guide explains.
- Responding in multiple languages. For businesses serving Metro Vancouver’s bilingual customers, a chat agent that moves naturally between English and Chinese covers a gap most small teams can’t staff. Just remember where those conversations often happen: in our experience, many Chinese-speaking customers ask on WeChat rather than in a website chat window, and a website bot never sees that inbox. Our WeChat private traffic playbook covers that channel.
- Qualifying enquiries before handoff. “Is this for a wedding or a corporate event? What date? Roughly how many people?” — collected before a human ever picks up the thread, so your first reply is a useful one.
If your enquiry pattern is dominated by this routine layer, a chatbot is doing real work.
Where they fail — and damage trust
The failures aren’t edge cases. They’re structural, and they cluster in four places:
- Complex or emotional situations. A customer with a complaint, a cancellation dispute, or a special request doesn’t want a fluent paragraph. They want a person. A bot that keeps politely paraphrasing while someone gets angrier is actively making things worse.
- Anything requiring judgment or exceptions. “Can you squeeze us in?” “Can I return this two days past the deadline?” These are decisions, and a bot either refuses rigidly or grants things it shouldn’t.
- Hallucinated policies. This is the expensive one. An ungrounded AI agent will confidently invent discounts, refund terms, and availability you never offered — and the customer holding a screenshot of that invented promise is now your problem.
- Dead-end loops. A bot with no human escape hatch traps people who repeat their question, get the same non-answer, and leave — often for good, sometimes via a one-star review that mentions your chatbot by name.
One bad bot conversation can undo the goodwill of many good ones, because people forgive a slow human far more readily than a confident machine that wasted their time.
A BC tribunal has already ruled on a chatbot’s mistake
The invented-policy risk isn’t hypothetical, and a widely reported Canadian case was decided right here in BC. In November 2022, after a death in the family, a passenger used the chatbot on Air Canada’s website to ask about bereavement fares before booking a flight from Vancouver to Toronto. The bot said they could submit a ticket for the reduced rate within 90 days of it being issued. The airline’s own bereavement page said the policy didn’t apply once travel was complete, and Air Canada refused the refund.
At BC’s Civil Resolution Tribunal, Air Canada argued it couldn’t be held liable for information provided by its chatbot. The tribunal member called that “a remarkable submission”, wrote that it “should be obvious” the airline is “responsible for all the information on its website”, and found negligent misrepresentation. The decision also noted that Air Canada never explained why customers should have to double-check one part of its website against another. Air Canada was ordered to pay $812.02 in damages, interest and fees (Moffatt v. Air Canada, 2024 BCCRT 149).
The amount was small. The lesson isn’t: whatever your bot says, you said, and a link to the correct policy page next to the wrong answer didn’t get the airline off the hook. It’s a small-claims tribunal decision, and CRT members aren’t bound by earlier CRT rulings. This is general information rather than legal advice, but it’s the right working assumption for any business putting a bot in front of customers.
The decision framework
Skip the vendor demo and answer three questions with real numbers:
1. How much of your enquiry volume is repetitive?
Read your last month of messages, DMs, and calls. If the same ten questions dominate, automation has something to automate. If most enquiries are genuinely varied, a bot will mostly be in over its head.
2. What share arrives after hours?
Check the timestamps. A significant after-hours share — common for restaurants, clinics, salons, and trades — is the strongest argument for a bot, because those are leads you’re currently losing outright.
3. What does answering cost versus what the bot costs?
Count staff hours spent on routine replies, then compare honestly against the full cost of a bot: setup, the monthly fee, and — the part every pitch omits — ongoing monitoring. An unwatched chatbot is a liability, not an asset. If nobody on your team will own it, don’t install it.
If you do it: five non-negotiable rules
- Ground it strictly on your real content. The bot should only answer from your actual FAQ, policies, and service pages — and say “I don’t know, here’s how to reach us” for everything else. This is the single control that prevents invented policies. It’s the same rule we apply to AI-assisted content: AI can draft, but anything it asserts that you can’t verify doesn’t go out.
- Always offer a clear path to a human. Visible from the first message, not buried three menus deep. The bot’s job is to handle the routine layer, not to hold the door shut.
- Disclose that it’s AI. People don’t hate talking to bots; they hate being tricked into it. Honest labelling is also where our thinking on AI content boundaries lands for every customer-facing use of AI. Canada’s federal, provincial and territorial privacy regulators, BC’s included, say the same in their joint principles on generative AI: when a tool is public-facing, people should know they’re interacting with AI and be told the privacy risks (Office of the Privacy Commissioner of Canada).
- Review the transcripts weekly. This is quality control — you’ll catch wrong answers before they compound. It’s also a goldmine: transcripts show you the exact words customers use to describe what they want, which is better marketing language than any brainstorm will produce.
- Treat the transcripts as customer data. Chat logs hold names, phone numbers and sometimes health or family details. For most BC businesses, that information falls under the province’s Personal Information Protection Act, which applies to private-sector organizations handling personal information of people in BC (OIPC BC). Have the bot collect only what you need to follow up, decide how long you keep the logs, and ask your vendor whether conversations are used to train its models. The regulators’ principles cover all three: limit collection, set retention schedules, and expect providers to disclose secondary uses such as training.
The alternative most small businesses should try first
Here’s the honest part vendors won’t tell you: a genuinely good FAQ page plus quick-reply templates for your inbox solves most of the same problem for free. Clear answers to your real top-ten questions, organized and linked prominently, handle the routine layer with zero risk of hallucination. Templates make your human replies fast without making them robotic.
And a strong FAQ page pays a second dividend: it’s exactly the kind of structured, factual content that search engines rank and AI assistants quote, the core of what our GEO guide covers. When someone asks a chatbot to recommend a business like yours, being quotable is what gets you recommended. Build that foundation first — if enquiry volume still overwhelms you afterwards, you now have the grounding content a chatbot needs anyway, and you can add one from strength.
Not sure which side of the framework your business lands on? Get in touch. If the answer is a better FAQ page, that’s part of our web design work.
Frequently Asked Questions
Is an AI chatbot worth it for a small business?
It depends on three numbers: how many repetitive questions you get, what share arrive after hours, and what answering them costs in staff time versus the cost of setting up and monitoring a bot. High volume of routine questions plus significant after-hours enquiries is a genuine fit. Low volume or mostly complex questions means a good FAQ page will serve you better for free.
What do AI chatbots get wrong for small businesses?
They fail at anything requiring judgment: complaints, exceptions, emotional situations, and edge cases. Worse, an ungrounded bot will confidently invent policies you never had — discounts, refund terms, availability — and a bot with no human escape hatch traps frustrated customers in loops. Both failure modes cost more trust than the bot saves in time.
What should a small business try before installing a chatbot?
A genuinely good FAQ page and a set of quick-reply templates for your inbox. Together they solve most of the same problem — routine questions answered instantly — at zero cost and zero risk of a bot inventing answers. A well-structured FAQ page also feeds SEO and makes your business easier for AI assistants to quote and recommend.
Am I legally responsible for what my chatbot tells customers?
Plan as if you are. In Moffatt v. Air Canada (2024 BCCRT 149), BC's Civil Resolution Tribunal rejected Air Canada's argument that it could not be held liable for its chatbot, said it should be obvious the airline is responsible for all the information on its website, and ordered it to pay $812.02. It is a small-claims decision and CRT members are not bound by earlier CRT decisions, but the reasoning is simple enough that a small business should assume its bot's answers are its own. This is general information, not legal advice.
Keep reading
Get in touch