A customer types her question into a bike shop's chat bubble: "Do you tune up bikes bought second-hand?" The answer lands, instant and beside the point: "Discover our maintenance packages here." She rephrases. Same answer. She types "talk to a human". The bubble offers her... the maintenance packages. She closes the tab. That same evening, in another small business in Metro Vancouver, forty messages wait for a reply that will go out at 10 p.m., too late for half of them. Two ways to lose the same customer. Between the two sits a well-mannered chatbot: a conversational bot, meaning a program that answers your customers in writing, inside a chat window, from answers you have entrusted to it. With AI, it no longer recites word for word, it rephrases. What remains is deciding what to hand over, what to keep, and where to put the exit door. That is exactly the plan of this article.
Two numbers that seem to contradict each other
On one side, your customers never all sleep at once: according to Zendesk's CX Trends 2026 report, 74% of consumers expect customer service to be available 24/7. On the other, they still distrust the machine: in 2024, Gartner measured that 64% of customers would rather companies did not use AI in customer service at all. A contradiction? Not really. The Gartner survey published August 4, 2026 (3,566 customers polled in February and March) reconciles the two: half of customers now find their interactions easier when a company uses generative AI, yet 87% consider it essential to be able to reach a human. People do not hate the bot. They hate the wall.
Meanwhile, Canada is gearing up. According to Statistics Canada, 19.2% of Canadian businesses were using AI in the second quarter of 2026 to produce goods or deliver services, three times the 2024 figure (6.1%). Among those users, the virtual agent already ranks third among the most common applications (28.2%), right behind data and text analysis. The detail that stings: among businesses with 100 or more employees that use AI, one in two (50.5%) has already hired its conversational bot. Your bigger competitors answer at night. The good news: a small business can do the same without losing its personality, provided it draws a clear line.
The dividing line
Everything is decided here. A chatbot earns its keep when the answer is stable and verifiable; it damages the relationship the moment the conversation calls for an ear. Here is the line as I draw it for a service or retail small business:
| The situation | Who answers | Why |
|---|---|---|
| Hours, address, parking, cancellation policy | The chatbot | The answer never changes, and nights and Sundays exist. |
| Order status, booking or moving an appointment | The chatbot, connected to your tools | A precise, emotion-free task, executed without tired mistakes. |
| Pointed product question, simple quote | The chatbot first, a human copied in | The bot prepares the ground; the human confirms the nuance in the morning. |
| Complaint, refund, rising tone | A human, fast | Emotion calls for an ear, and goodwill gestures call for judgment. |
| Vague request, edge case, negotiation | A human | Two off-target replies are enough to lose the person for good. |
One more reason to polish the "chatbot" column: your customers already ask their questions to AIs you do not own. According to a second Gartner survey from July 2026, in their most recent service interaction, customers were about three times more likely to consult ChatGPT, Gemini or Copilot than the company's own chatbot. Your bot is the only place where you write the answers. They had better be good, and they had better live on your website too: I have shown elsewhere how a well-built FAQ page gets cited by AIs while it serves your customers.
Three rules that protect the relationship
Gartner's August survey contains the instruction manual, and it fits in three rules:
- Announce the machine. No fictional "Julie" with a stock photo. A bot that introduces itself as a bot starts with trust; a bot in disguise loses it twice. Say what it can do ("I answer questions about hours, prices and appointments") to set expectations from the first line.
- The human door, one tap away. Among customers who refuse AI, the first factor that would change their mind is the ability to switch to a human when needed. The "talk to someone" button must be visible from the first screen, with no secret password, and it must lead somewhere: an inbox checked at set times, or a callback promised by a stated hour.
- When in doubt, hand it over. Gartner puts it plainly: the bot must never be a mandatory gate. Set it to transfer after a single failure to understand, passing along the history so the person does not have to repeat everything. Nobody has ever enjoyed retyping their question to a machine that apologizes beautifully.
These rules are not decorative politeness. The same survey observes that customers forced through several failed AI exchanges before reaching a person stop using the tool afterwards. A badly tuned bot does not just score zero: it digs.
The real cost in 2026
Three pricing models dominate, from simplest to strongest. First, live chat without AI: often already included in your current tools, it puts a window on your site and forwards the messages; an honest start. Next, monthly subscriptions, a few dozen dollars, where the bot answers from your content. Then billing by outcome: Intercom's Fin, the best-known example, charges US$0.99 per settled outcome (a resolved answer, a clean handoff or a disqualified request), with a 50-outcome monthly minimum, so a floor of about US$49.50. The vendor's published case studies and independent tests place real resolution rates between 40 and 50% of conversations: for a small business receiving 150 a month, that means 60 to 75 resolutions, so between US$59.40 and US$74.25 monthly. Less than one evening a week spent answering the same five questions.
Two invisible expenses complete the bill. Writing time first: the bot is only worth the answers you give it, so budget a few hours at the start and one review per season. Caution second: Statistics Canada notes that cybersecurity and privacy concerns are the top barrier for Canadian businesses facing AI (13.4%), ahead of cost (10.6%). The hygiene rule is simple: give the bot only what you would publish on your website, and keep customer files out of its reach until your vendor has put its guarantees in writing.
Start with your twenty answers
Here is the cheapest, sturdiest path I know. Reread a month of messages and emails. Note every question that keeps coming back: there are rarely more than twenty. Write a clean answer for each one, in your voice, with your real prices and timelines. Publish them on an FAQ page of your site. At this stage you have paid nothing, your customers already find their answers at midnight, and the AIs that talk about you have a clean source to cite. The chatbot then becomes the second step, not the first: you plug it into proven answers instead of letting it improvise. For 30 days, track three numbers: the share of conversations settled without you, the number of "talk to a human" requests, and bot-related complaints, which must stay at zero. If that third line moves, tighten the dividing line. And if your real issue is not answering questions but following up on time, your CRM's AI-free plumbing will often do better than a chatty robot.
Frequently asked questions
How much does an AI chatbot cost for a small business?
From zero to about a hundred dollars a month for most small businesses. Simple live chat is often already included in the tools you have (website, CRM). AI agents billed by outcome, like Intercom's Fin, charge US$0.99 per conversation actually settled, with a 50-outcome monthly minimum, so a floor of about US$49.50. For a small business receiving 150 conversations a month where the bot settles 40 to 50% of them, the bill lands between US$59.40 and US$74.25 a month. The real cost sits elsewhere: the answers you have to write, then keep up to date.
Can an AI chatbot replace a small business's customer service?
No, and your customers say so themselves: 87% consider it essential to be able to reach a human when a company uses AI for customer service, according to Gartner's August 2026 survey. The chatbot shines on stable questions (hours, base prices, order status, appointments) and works nights without sighing. Complaints, edge cases and goodwill gestures stay human: that is where loyalty is won. The right question is not whether to replace, but which half to hand over.
Where do you start with no budget and no tool?
With your twenty answers. Reread a month of messages, note the questions that keep coming back, write a clean answer for each and publish them on an FAQ page. It costs nothing, it serves your customers by tomorrow morning, and the day you plug in a chatbot, its manual will already be written. Then track three numbers for 30 days: the share of questions settled without you, the number of people asking for a human, and bot-related complaints, which must stay at zero.
Curious about what your business should hand to a chatbot, and what it should keep?
Book a discovery call