AI assistants for small business

AI assistants for small business are good at answering routine questions, drafting text, sorting enquiries and working at 2 a.m. without complaint. They are bad at anything that needs judgment, accountability or facts they were never given. Used well, an assistant takes the repetitive part of customer service off your plate and makes sure no enquiry waits until morning. Used carelessly, it invents prices, promises things you do not offer and puts your name on the mistakes.

This article is about AI assistants that talk to your customers, typically on your website or in messengers, and about the internal tools you might use yourself. The principles are the same: know what the tool is for, give it accurate material and keep a human in charge.

What an AI assistant actually is

A modern AI assistant is built on a large language model: software that has learned patterns of language from huge amounts of text and can produce fluent answers to almost any question. On its own, it knows nothing specific about your business. To be useful, it needs a knowledge base (your services, prices, policies, hours, service area) and instructions about tone, limits and when to hand over to a person.

The key thing to understand: a language model produces plausible text, not verified truth. When it lacks information, it may fill the gap with something that sounds right. That is why the setup matters more than the model.

Six jobs AI assistants do well

1. Answering repeated questions. Opening hours, parking, what is included in a service, whether you work in a certain town, how to prepare for an appointment. If the answer is in your knowledge base, the assistant handles it reliably and instantly.

2. Collecting details before a human steps in. Name, phone, postal code, type of job, photos, preferred dates. The assistant can ask follow-up questions naturally, so the lead arrives ready to quote.

3. Working after hours. For many local businesses, a large share of enquiries come in the evening when people finally have time. An assistant keeps those conversations alive until you are back.

4. Booking standard appointments. Connected to a calendar, an assistant can offer slots and confirm a booking. See automating bookings and reminders for the wider setup.

5. Speaking several languages. Many models handle French and other languages well, which helps businesses serving bilingual or immigrant communities. Answers in each language should still be checked by someone who speaks it.

6. Drafting for you. Internally, AI tools can draft replies, newsletters, job descriptions and summaries of long email threads. You review and send. That alone saves owners hours a week.

Five things they should not do

1. Quote custom prices. If the price depends on a site visit or measurements, the assistant should collect details and promise a human follow-up, not invent a number.

2. Give medical, legal or financial advice. Clinics, law firms and financial advisors should keep the assistant to logistics and general information, with a clear handover for anything personal.

3. Make commitments you have not approved. Refunds, warranties, exceptions to policy. These need a person.

4. Handle complaints alone. An upset customer talking to a bot gets more upset. Route complaints to a human quickly.

5. Pretend to be human. Be clear that customers are talking to an assistant. It builds trust and avoids awkward moments when it makes a mistake.

The accountability question

In Canada, there is a well-known 2024 decision in which a British Columbia tribunal held an airline responsible for incorrect information its website chatbot gave a customer about a fare policy. The business could not shift the blame to the bot. The lesson for small businesses is simple: whatever your assistant says on your website is treated as your business speaking. Keep it to information you are willing to stand behind, and review conversations regularly.

This is a general point, not legal advice. If your assistant will deal with regulated topics, ask a lawyer how your industry's rules apply.

Privacy and consent in Canada

An assistant collects personal information, so PIPEDA applies, and for customers in Quebec, Law 25 adds stricter requirements, including transparency about automated processing in some cases. Practical steps:

• Mention the chat assistant in your privacy policy and say what happens to conversations.
• Collect only what you need to help the customer.
• Check the vendor's data storage location and retention settings.
• Do not feed sensitive personal data into tools that may use it for training.

My articles on what your privacy policy needs and Quebec Law 25 cover the details.

How to start without regret

1. Pick one job. For example, answering the top 15 questions and collecting lead details after hours.
2. Write the knowledge base. One clear document with current facts. If your website has outdated pages, fix those too.
3. Set the limits. List topics the assistant must not answer and what it should say instead.
4. Define the handover. Who gets notified, through which channel, and how fast they respond.
5. Test with real questions. Use actual emails and messages from the last few months.
6. Review weekly at first. Read conversations, fix wrong answers, add missing information.

Signs your assistant needs attention

Once it is live, a few warning signs tell you the assistant needs work: customers repeat the same question in different words, conversations end abruptly after the bot's answer, staff hear "but your chat said..." on the phone, or handovers pile up without a reply. Each of these points to a missing fact, an unclear instruction or a gap in the handover. Fix the source, not just the single conversation.

What it costs

A professionally built chatbot from me costs $2,500, and the Leads under control package (CRM setup plus chatbot) costs $5,900, with platform and AI usage fees on top, paid directly by you. More detail in how much a website chatbot costs. I deliver chatbots and AI assistants with a trusted partner; the contract, deadlines and communication stay with me, and I handle the CRM side myself so every conversation lands with an owner.

If you want to see the options, my AI assistant page describes what a typical project includes. The honest summary: an AI assistant is an excellent junior team member for repetitive work. It is not a manager, a lawyer or a salesperson for complex deals, and it should never be left alone with your reputation.