Knowledge base

AI agent or chatbot: what do you really need?

In short, a chatbot talks and an AI agent acts. A chatbot answers questions based on your information. An AI agent sets steps in motion itself and carries out actions in your systems. It judges a case and acts on it. Choose a chatbot for frequently asked questions, and an agent when each case needs to be judged and acted on.

The short answer.

The difference is in what the thing ultimately delivers. A chatbot gives an answer. Someone asks a question, the bot looks it up in your information and types a reply. Nothing more happens, and that's often exactly enough. An AI agent goes a step further. It reads a case, decides what needs to happen and then actually does it in your systems, for example creating a booking, routing a ticket or preparing a draft. A chatbot is someone to talk to, an agent is someone who gets things done.

The practical choice follows from that. If it's the same questions over and over and a good answer exists for them, a chatbot is enough, and it's live sooner and cheaper. If each case needs to be judged and then something really has to happen in another system, you need an agent. The tricky part is that almost everything gets called an agent these days, including a simple chatbot with a new label. Further down I explain how to see through that label, and what each one costs in practice.

Side by side.

This is the difference on the points that matter when you choose. The first column is the criterion, and next to it is what a chatbot does and what an AI agent does. The cost figures are an indication based on my own prices, excluding VAT, and are listed in full on the price list.

CriterionChatbotAI agent
What it doesAnswers questions based on your informationJudges a case and carries out actions in your systems itself
When you choose itRecurring, frequently asked questionsEach case needs to be judged and acted on
Works with your systemsReads along, doesn't write anything itselfWrites and acts in your systems, through integrations
Needs a person involvedHardly, it just keeps talkingYes for important steps, a person has the final say
ComplexityLow, live quicklyHigher, needs integrations and monitoring
Cost indicationOften a quick win from €750A smart system from €3,500
ExampleAnswers about opening hours and ratesAI receptionist that handles the question and passes it on

When a chatbot is enough, and when you need an agent.

Start with one question. What needs to happen after the answer? If the answer itself is the end point (someone wanted to know something and now they know it), a chatbot is the right tool. Think of opening hours, rates, how a membership works or where someone can find a form. The same questions, time after time, with answers you know in advance. A chatbot built on your own information handles that without anyone having to step in, and it's often a real quick win from €750.

You need an agent as soon as the answer is only the beginning. A case comes in, it has to be judged, and based on that something really has to happen. Recording a reservation, sending a complaint to the right person, preparing a draft that someone only needs to check. That's judging and acting, not looking things up and answering. An agent like that talks to your other systems through integrations, so it's more likely to fall under a smart system from €3,500. Decisions that matter need a person in the loop, because an agent is never as predictable as a fixed rule.

In practice the two often overlap. The AI receptionist at Inge de Bruijn Zwembad started out answering customer questions, 146 of them in the first period, and that's chatbot work. It becomes agent work the moment a question can't simply be answered and something has to be set in motion or passed on. You can read how that's built up in the AI receptionist case study, and what an AI assistant on your own knowledge can do for you is on the service page.

Watch out for agent-washing.

Ever since agent became the buzzword, almost everything goes by that name. A chatbot in a sharp suit gets called an agent, so does a script with a few if-this-then-that rules, and the price goes up with it. That's agent-washing. The label changes, the work underneath doesn't. I find that annoying, because it makes it impossible for a business owner to compare apples to apples.

The test is simple. Ask whether this thing carries out actions in your systems itself, or whether it only talks. Can it judge each case and then act, or does it keep giving the same kind of answer? Can it break something if it goes wrong, or is the damage limited to a wrong answer? If the answers point toward talking and answering, it's a chatbot, whatever it's called, and it should come with a chatbot price.

A chatbot not being an agent doesn't make it any less. For many businesses a chatbot is actually the better choice, since it's live sooner, cheaper and more predictable. You only build an agent where things really need to be judged and acted on. If you're not sure where your question fits, that's exactly what I work out per process in the free process review, without selling you an agent when a chatbot will do.

Chatbot or agent: what fits your process?

In the free process review I look at each process to see whether a chatbot will do or whether an agent pays for itself, and roughly what it costs, with a fixed price up front. I won't sell you an agent if a chatbot is better for you. That's the deal.

No newsletter spam. At most one email a month, and unsubscribing is just a reply.

Walk me through one process