Conversational assistants based on artificial intelligence are spreading across business websites. Behind the chat window that opens at the bottom of the page, actual uses vary a great deal: some assistants genuinely help visitors, others complicate the journey. This article takes stock of what an AI chatbot can bring to the website of a small or medium-sized business, of the prerequisites that are often underestimated, and of the limits to know about before getting started.
What exactly are we talking about?
The word “chatbot” covers two rather different families of tools. The first is scripted chatbots: they follow a decision tree written in advance (“Choose an option: quote, opening hours, contact…”) and do not understand freely worded questions. The second is assistants based on generative AI: they interpret a question in natural language and produce an answer from a knowledge base they are given, for example the pages of the website, an FAQ or internal documents.
| Scripted chatbot | Generative AI assistant | |
|---|---|---|
| Understanding | Predefined choices only | Questions in free language |
| Answers | Texts written in advance | Generated from a knowledge base |
| Set-up | Writing the scripts | Preparing and maintaining the knowledge base |
| Main risk | Rigid journey, dead ends | Inaccurate answers if the base is incomplete |
The uses where an AI assistant is genuinely helpful
On the website of a small or medium-sized business, a conversational assistant is mainly justified when a significant volume of repetitive requests comes in by phone or email. The most common use cases:
- Answering frequently asked questions outside opening hours: indicative prices, lead times, service area, how to book an appointment.
- Directing the visitor to the right page of the website: a service page, a contact page, a quote form.
- Qualifying a request before contact is made: gathering the need, the context and the contact details, then passing a summary on to the team.
- Helping with specific procedures: order tracking, preparing for an appointment, documents to provide for an application.
In these situations, the assistant does not replace a human: it handles the simple requests and passes on the others, which reduces the number of back-and-forths before a first useful exchange.
The prerequisites that are often underestimated
An AI assistant does not produce good answers by magic: it gives back what it has been given. Three prerequisites determine the quality of the result.
- A clean, up-to-date knowledge base: if the information on the website is incomplete or contradictory, the assistant will be too. Preparing this content is often the bulk of the work.
- A defined scope: specify what the assistant is allowed to handle, and plan a systematic hand-off to a human (form, email, phone) for everything else.
- Regular monitoring: re-read samples of conversations, correct inaccurate answers, add to the base when new questions appear.
Limits and points to watch
Several limits need to be understood before deploying an AI assistant on a website.
- The risk of an inaccurate answer: a generative model can confidently state a wrong answer, particularly when the question falls outside its knowledge base. On matters with real consequences (firm prices, contractual terms, regulatory advice), it is wiser to refer the visitor to a human rather than let the assistant answer.
- Data protection: conversations may contain personal data. The GDPR requires you to inform visitors, to limit collection to what is necessary, and to know where that data is stored and for how long.
- The recurring cost: beyond the initial set-up, an AI assistant involves a subscription or usage-based billing, plus the time spent maintaining the knowledge base.
- The effect on the experience: a chat window that opens by itself, hides the content or insists on starting a conversation degrades the journey instead of improving it. The assistant should remain discreet and available, not intrusive.
A useful AI assistant answers correctly, knows its limits and hands over to a human at the right moment. These three qualities are built, they cannot be improvised.
How do you know whether it makes sense for your website?
A few simple questions help you decide. Do you receive repetitive requests every week that the website already answers, or could answer? Do your visitors look for information outside your opening hours? Do you have the time, in-house or through a service provider, to maintain the knowledge base? If the answer to all three questions is yes, a conversational assistant is worth looking into. Otherwise, a well-structured FAQ and a clear contact form often provide the same service, at a far lower cost.
Wondering whether an AI assistant would be worthwhile for your website or application? Let's talk: an initial conversation helps assess whether the need justifies it and where to start.