"Should we have a chatbot" is the wrong question today. The better one is: what should it do, and who is responsible for what it tells your clients. That decides whether a tool for a few dozen zloty a month is enough, or whether you need your own agent built on your knowledge base.
In this article we compare three realistic options, what they cost, how to choose between them - and what none of them will do.
Short answer: for most small businesses an off-the-shelf AI tool is enough to start: it costs from a few dozen to a few hundred zloty a month and can be live in hours. A custom agent makes sense when you need integration with your own data, control over the answers and consistency with your brand, and one client is worth enough to justify it. A rule-based chatbot - the "press 1" decision tree - only holds up for very simple questions.
Kluczowe wnioski
- An off-the-shelf AI tool is enough to start for most small businesses - you can launch it in hours.
- A custom agent makes sense with integrations, control over answers and higher client value.
- Bills usually get out of hand through per-seat pricing and limits on AI conversations.
- Track one number: how many conversations end in contact, a booking or an order.
Three kinds of website chatbots
| Option | How it works | Approx. cost | When to choose it |
|---|---|---|---|
| Rule-based widget | Answers only foreseen questions, sends the rest to a form | usually included in a tool's free plan | simple FAQs and low traffic |
| Off-the-shelf AI tool | Learns from your content, talks naturally, hands harder cases to a person | usually about $20-30 per seat a month, the AI module often priced separately | most small and mid-sized businesses |
| Custom AI agent | Built on your knowledge base, wired to your calendar and CRM | a build cost plus conversation costs or a subscription | integrations, control over answers, brand |
In Poland you will most often come across Tidio and Smartsupp - both have free plans and separately priced AI modules - and internationally Crisp and Intercom. Prices change often, so treat these numbers as an order of magnitude and check the current price list before deciding. Watch two places where bills usually get out of hand: per-seat pricing as the team grows, and a monthly cap on conversations handled by AI.
When an off-the-shelf tool is enough, and when you need your own agent
An off-the-shelf tool is enough when
- The questions repeat: opening hours, prices from the price list, availability
- The team is small and you want to start this week
- You are still checking whether clients use chat at all
- You want a predictable monthly budget
Go custom when
- The agent has to use your price list, calendar or CRM
- Answers must match your brand and sales process
- You need control over what the agent says and refuses to say
- One client is worth enough that the quality of the conversation matters
An off-the-shelf tool has one more advantage: it shows what clients actually ask. Two or three months of conversations give you a ready list of questions and a ready knowledge base if you later decide to go custom.
Before you deploy: five things to settle
- The goal: one number that should improve.
- The knowledge base: where answers come from and who keeps it current.
- Escalation: when a person takes over, and how fast.
- Data: who processes it, where it is stored and whether it trains models.
- Disclosure: a clear message that the client is talking to an AI.
The most common mistake is deploying without an answer to the first point. A bot that simply "had to be there" usually ends up answering questions nobody asks, and nobody can say whether it pays off.
What a chatbot will not do
- It will not bring you clients. It answers the people already on your site. If there is no traffic, the problem is higher up - in visibility, not in service.
- It will not be infallible. Hallucinations can be greatly reduced with a narrow knowledge base, a plain "I do not know" and escalation to a person, but not eliminated.
- It will not replace a salesperson in complex sales. It can take over the first contact and qualification.
- It cannot advise where professional responsibility is involved. In a law firm or a clinic an assistant's role ends at organisational matters - we cover this in a law firm website and an aesthetic clinic website.
How to check whether it pays off
One number is enough to start: how many conversations end in contact, a booking or an order. Add two supporting ones: the share of conversations handed to a person, and the list of questions the bot could not answer - ready material for improvements. Review the transcripts once a month; that first review usually does more than changing the model.
Chat alone is not the whole of automation either. If enquiries get lost or nobody follows up, tidying the process will change more than a widget - we break it down in Sales automation for small business.
Chatbots and site speed
Every widget is another script, and they can weigh hundreds of kilobytes. Load it only after an interaction or with a delay, and check the result on a phone - speed translates into the number of enquiries. Make sure the chat window does not cover the content on a small screen and is easy to close.
When it is worth building your own agent
When it has to do things an off-the-shelf tool cannot: use your price list and calendar, hold a conversation in your brand's tone, save a lead in your system with the full context. What such a deployment looks like and what it costs is covered in An AI agent on your website. If you are still planning the budget for the site itself, start with How much does a website cost.
Frequently asked questions
Will a chatbot hurt my Google rankings?
Not in itself. What can hurt is a heavy script that slows the site down and a window that covers the content on mobile. A widget loaded only after an interaction is safe in that respect.
Do I have to disclose that the client is talking to an AI?
Yes. Since 2 August 2026, AI systems that converse with people must clearly state that they are AI, unless it is obvious from the context.
Is client data safe?
That depends on the provider, not the technology. Establish who processes the data, where it is stored, whether conversation content is used to train models and whether you have a data processing agreement in place. Until you know, do not collect sensitive data through chat.
How long does it take to launch?
An off-the-shelf tool: from a few hours to a few days. A custom agent: usually around three weeks, depending on the knowledge base and the number of integrations.
Will a chatbot replace the contact form?
No. Some people prefer a form or a phone call, especially for a serious matter. Chat is an additional route, most useful in the evening and at weekends.
Summary
Choosing a chatbot is not a technology choice but a decision about how much control over the conversation you need. Start with an off-the-shelf tool, collect your clients' questions, and move to your own agent when the limits start costing more than a build.
If you want to talk it through with your own numbers, write to us.