A Chatbot for a Small Business: What Works and What Is Just a Gimmick

Customers message you after 6 p.m. - on chat, Messenger, email. You reply in the morning, and some of them have already bought elsewhere. That is exactly why a chatbot for a small business is no longer a big-corporation gadget. The catch: the market promises miracles, and half-baked deployments end with a bot that irritates customers and solves nothing. In this post we show what really works in a small company, what a bot should never do, and what it roughly costs.

A handful of questions generates most of the traffic

Before you think about technology, go through your inbox and chat history. The same pattern repeats in most companies: a small pool of topics, usually around one fifth, accounts for the vast majority of messages. These are the questions worth automating:

• order status - "where is my parcel, my job?",
• pricing and scope - "how much does it cost and what exactly is included?",
• availability and dates - "do you have a free slot or the product in stock?",
• practical matters - opening hours, directions, returns, payments.

The rest - negotiations, unusual problems, difficult emotions - should still be handled by a human. A chatbot is one piece of a bigger puzzle we describe in our post on process automation in a small company.

A chatbot for a small business: three maturity levels

An FAQ bot is a good start when the questions are predictable. A knowledge-base bot uses a language model plus your materials: price list, terms, service descriptions - it answers naturally, but only based on what you feed it. An agent additionally connects to your systems through APIs, so instead of generalities it gives specifics from the database - how such connections work is explained in our post on system integration in a company. We deliver all three levels as part of our AI solutions.

What a chatbot should never do

• Promise prices and discounts - a model can get carried away; binding offers come from a human.
• Close complaints without a human - an upset customer plus a bot is a recipe for escalation.
• Pretend to be human - the customer should know they are talking to a machine.
• Collect sensitive data - a bot should never ask for more than the case requires.

The iron rule: every conversation must offer a one-click handover to a human, and the bot should escalate on its own whenever it is unsure. AI supports, it does not replace - the same way we see it in programming, as we wrote in why AI will not replace programmers.

Implementation: tidy knowledge beats model choice

The most common reason for a weak bot is not "a bad AI model" but messy company knowledge. A sensible plan looks like this:

• write down the 20-30 most frequent questions from email and chat,
• tidy up the sources: current price list, terms, FAQ and procedures in one place,
• set the boundaries: what the bot answers and what it hands over straight away,
• run a pilot, e.g. only on the offer page, and watch the conversations,
• review transcripts every week and keep filling the knowledge base.

Measure the results: the share of conversations resolved without a human, time to first answer, and whether customers abandon the chat halfway through.

What does running a chatbot cost?

Beyond implementation, plan for running costs. The language model is billed by usage - with low traffic it is usually in the range of tens to a few hundred PLN per month. Add updating the knowledge base and regularly reviewing conversations. Care for the bot, hosting and small fixes can be covered by a subscription from PLN 500/month - just like maintenance of any other application.

Where to start?

Not with picking the technology, but with the list of customer questions. If most of them fit simple scenarios, an FAQ bot for a few thousand PLN is enough. If customers mostly ask about order status or free slots, aim for an agent connected to your systems. Want to check which level makes sense for you and what it would cost? Describe your situation in our quote calculator - you will see a ballpark price range on the spot, get a first reply within 24 hours, and receive a binding quote after a short call.