A chatbot is software that holds a conversation in text or speech to answer questions or complete a task. Older chatbots follow a decision tree the builder drew; modern ones use a language model and can handle questions nobody scripted. The practical difference for a business is whether it can only answer, or can also act in your systems.
A chatbot is software that holds a conversation — in text or speech — to answer a question or complete a task. The label covers two quite different things: older bots that follow a decision tree someone drew in advance, and modern ones built on a language model that can handle questions nobody scripted.
For a business the meaningful distinction is narrower than the marketing suggests. It is not rule-based versus AI. It is whether the bot can only answer, or can also act — look up the order, change the appointment, raise the ticket — in the systems you already run.
A rule-based bot matches what the visitor typed against patterns its builder anticipated and follows a fixed path. It is predictable, cheap and brittle: anything off-script produces "I didn't understand that". An AI chatbot passes the question to a language model, so it copes with phrasing nobody planned for and with two questions in one sentence. The trade is control — it will answer things you never approved unless you constrain it deliberately.
The honest answer is: deflect the questions your team answers most often, at the hours nobody is working. Opening times, order status, delivery, warranty scope, appointment changes, the same five product questions. That is unglamorous and it is where the payback is, because those questions are high-volume and low-value. Everything beyond it — negotiating, judging an exception, calming an angry customer — is still better handed to a person, and the bot's job is to hand it over cleanly.
Three cost lines, and only one of them is the build. There is the build (scope, content, integrations, testing), the running cost (model usage, hosting, platform fees) and the maintenance nobody budgets: keeping answers true as products, prices and policies change. A bot that was correct in March and unmaintained by September is worse than no bot. We publish fixed figures for the build on what a chatbot costs.
A chatbot converses; an agent acts. Given "move my appointment to Thursday", a chatbot tells you how to do it, while an agent has permission to call the scheduling system and do it, then report back. Agents need something a chatbot does not: named, validated operations against your systems, and a clear rule about what they may do without a human confirming. That difference is set out in AI agent vs chatbot.
Three places, consistently. Stale content — the bot repeats a policy you changed. No exit — a customer who needs a human cannot reach one, which converts a small problem into a complaint. And confident invention when the bot has no grounding in your real data and fills the gap. All three are design decisions, not model limitations, which is why they are avoidable and why they keep happening.
It will if it blocks them. Customers tolerate bots that are fast and honest about their limits, and resent bots that trap them.
Yes. Current language models handle Dutch well, including the u-form and sector vocabulary, and can switch language mid-conversation.
Only for the questions that are actually about a customer. Most FAQ traffic needs no personal data at all, and keeping it that way simplifies your AVG position considerably.
Want this applied in your business? See how we take it to production:
We build this AI in production — at fixed prices, with one named expert. Start with a free consultation.
Book a free consultation →