“Chatbot” and “AI agent” are often used as if they mean the same thing. They do not have to. A chatbot usually answers questions in a conversation. An agent is normally given a goal, access to selected tools and permission to complete a sequence of actions, such as checking availability, creating a draft booking and asking a person to approve it.
For a small business, the useful choice is not the most impressive label. It is the smallest system that solves a real customer problem safely. A clear FAQ and a well-written contact form may be all you need. A carefully limited agent can be useful when a repeated process involves several steps and a team member has a clear way to review the result.
What a chatbot can do well
A chatbot can answer common questions about opening times, areas covered, delivery, parking, booking rules or the services you provide. It can point a visitor to the right page and collect a short enquiry when the alternative is leaving the site.
Give it a small, approved knowledge base. Use the same service descriptions, prices and policies that your team uses. Tell it what to do when the answer is missing: say it does not know, link to a person or collect a question for follow-up. A confident wrong answer costs trust quickly.
What an agent adds
An agent can interpret a request, choose from a set of allowed actions and pass information between tools. For example, a B&B agent might check the dates a guest mentions, explain the booking conditions and create a draft enquiry. An online store agent might look up an order status, identify a delivery exception and prepare a message for a staff member.
The agent needs boundaries. Decide which data it can read, which actions it can take automatically and which require approval. Sending a general acknowledgement is different from issuing a refund, changing a booking or making a legal or financial promise.
Choose the problem before the technology
List the questions and tasks that repeat every week. Note how long they take, where errors happen and what a good outcome looks like. If the task is simply finding a page, improve the navigation and write the answer clearly before adding a bot.
An assistant is more likely to help when the information is stable enough to maintain and the person asking expects a quick answer. It is less suitable when every job needs a site visit, an experienced judgement or a bespoke quote.
Make the handover easy
A visitor should be able to ask for a person without arguing with the assistant. Show the business phone number, email or booking option at the point of need. Pass the conversation to the team with the original question and any information already collected.
Tell the visitor what happens next. If a human replies during working hours, say so. If a message will be reviewed the following morning, set that expectation. The handover is part of the product, not a failure of the bot.
Keep your knowledge base current
Write down the sources the assistant may use and who owns each one. Review opening times, delivery rules, prices, service areas and availability when they change. Add a date to internal notes so an old promotion cannot remain in the answer forever.
Test the assistant with ordinary questions, misspellings, local place names and requests that combine two services. Ask it to explain when it is unsure. Keep a log of questions it could not answer; those are useful clues for a clearer website or a new FAQ.
Protect customer information
Avoid putting personal information into a tool unless you understand where it is stored and who can access it. Give the assistant only the access it needs. Do not let it reveal another customer’s order, appointment or contact details because somebody asks in a clever way.
Use a separate path for sensitive requests. A human may need to verify identity before discussing an account, changing a booking or dealing with a complaint. Make the rule visible to your team and test it regularly.
Measure usefulness rather than conversation length
Track questions answered, handovers, completed bookings and enquiries that became suitable opportunities. A long conversation can signal that the assistant is helpful, but it can also show that the person is struggling to get a direct answer. Ask a few customers whether the interaction saved time or made the next step clearer.
Compare the assistant with the normal route. If people are abandoning a form because the bot asks too many questions, remove the questions. If it resolves a repeated query and frees the owner to serve a customer, keep the improvement.
Start with a contained pilot
Choose one channel, one topic and a short review period. A website FAQ chatbot about services and opening times is easier to supervise than an agent with access to every business system. Give it a small set of approved replies, a human escalation route and a daily check during the pilot.
Keep a list of answers that need improving. Ask the person who actually handles the questions to review them, because they know the language customers use and the exceptions a generic setup misses.
When a simple page is better
If customers need photographs, detailed comparisons or a clear form, a well-structured page may serve them better than a conversation. If most questions are about information that changes often, publishing an accurate page and linking to it can be safer. Technology should make the answer easier to reach.
Hart & Forge helps UK SMEs decide where a chatbot, an AI-assisted workflow or a straightforward website improvement will have the most useful effect. Talk to us about a sensible AI automation pilot, and we can start with the process you want to improve.
