Chatbot development: when it works and when you should pick something else

At Your Tech Club in Amsterdam we build customer service chatbots for your website, WhatsApp chatbots and custom chatbots for your own team. If you run an international company with an office in the Netherlands, your chatbot will often have to work in English and Dutch at once. What a chatbot is and which types exist, you can read in What is a chatbot?. This page explains how we choose.

When is a chatbot enough?

A chatbot is enough when customers keep asking the same questions and the answer is already written down somewhere. Think of delivery times, opening hours, returns conditions or how a product works. The chatbot looks up the answer and gives it in plain language. It does not have to change anything in your systems, so it is quick to build and easy to manage.

A modern chatbot runs on a language model that only answers from your sources: your help pages, manuals and terms. That technique is called RAG, and I explain how it works on our glossary page about RAG. The advantage is that the chatbot makes up less. On top of that, you can see which source each answer comes from.

My rule of thumb is simple. A chatbot answers questions, an AI agent handles them.

A few questions where a chatbot works well:

• "Can I still exchange this jacket after 30 days?"
• "How many hours of leave do I build up on a 32-hour contract?"
• "Does this charging point work with an older car?"

In every case the answer is already in a document. It is just that nobody reads that document.

We built a simple form like this for a data and AI company: a contact form that answers by itself. The visitor describes in their own words what they want to achieve. A language model assesses that question and gives a substantive first answer straight away, before anyone from the company gets involved. It did not need an integration with an order system or CRM.

When do you need an AI agent instead of a chatbot?

You need an AI agent when the customer does not just want an answer but wants something done. Changing an order, registering a return or moving an appointment, for example. For that, the software has to look in your systems and change something there. A chatbot cannot do that. An AI agent can, with the permissions you give it.

Take the question "where is my order?". A chatbot will usually say that the tracking code is in your email. An AI agent looks up the order in Shopify or your ERP, sees that the parcel has not moved for three days and immediately creates a ticket for your team. That is a very different conversation.

You can read what an agent is in What is an AI agent?. If you want one built, see AI agent development. An agent needs integrations. We can connect any system; our integrations page shows how, from Exact Online (a widely used Dutch accounting package) to HubSpot.

When is a FAQ page better than a chatbot?

A FAQ page is better when you get few customer questions, when the answers are short or when your knowledge is not in order yet. A chatbot on a messy knowledge base gives messy answers. It just does so faster, and with more confidence. In that case, write ten good answers on one page first. Google can read that page too.

What I would not do: commission chatbot development because a competitor has one. That sounds silly, but it happens more often than you would think. First you need to know which questions come in and how much time they take. Without those numbers you build something you will never know is working.

I would also never put a chatbot in place as a gatekeeper. You know the kind: you want to talk to a person, but the bot sends you back to the same help page three times. A good chatbot recognises when it does not know and then hands the conversation over, with the question and the context attached. As a result, your customer does not have to type everything again.

Finally, a chatbot that nobody maintains goes stale quickly. If your returns policy changes, the chatbot should know the same day. If nobody is going to do that, start with a FAQ page instead.

A common outcome of the count: most questions are about a handful of topics, such as delivery times or returns. Then a better help page or a standard reply in the mailbox often solves it, and a chatbot only comes up later.

Which chatbot can we build for you?

Our chatbot development covers three types: a customer service chatbot on your website, a WhatsApp chatbot and a custom chatbot for your own staff. All three answer from your sources, hand a conversation to a person when needed and say from the first message that you are talking to AI.

Customer service chatbot on your website

This is the best-known form. The chatbot sits on your site, answers questions from your help pages and terms, and creates a ticket when it gets stuck. On AI customer service you can see how we approach this more broadly, from chatbot to helpdesk. Our AI employees for support show per task what is possible.

WhatsApp chatbot

Many customers would rather send a message than an email. A WhatsApp chatbot runs on Meta's WhatsApp Business Platform. According to Meta, you can use it for automated conversations and to hand a conversation over to a member of staff (WhatsApp Business Platform, accessed October 2026). If orders also need to come in through WhatsApp, our AI employee for WhatsApp order processing is the logical next step.

Custom chatbot for your own team

Not every chatbot needs to face outwards. Internally, a chatbot often works even better, because colleagues ask predictable questions. Think of an HR helpdesk for questions about leave, or an assistant for the IT service desk. A custom chatbot like this only looks in the documents that the employee already has access to. How strict can that be? See an AI assistant that pre-fills a form from the collective labour agreement, with the source behind every answer. If the answer is not written down anywhere, the field stays empty.

Chatbot development: how do we approach it?

We start by counting, not by building. First we look at which questions come in and how much time they take. Then we get the sources in order, build the chatbot and test it with real questions from your customers. Only when the answers are right does it go live, with a person behind the handover button.

In practice, it looks like this:

1. Collect the questions. We pull emails, tickets or WhatsApp conversations from the last few months and group the questions.
2. Check the sources. Are your help pages and terms still correct? If not, we fix those first.
3. Build and connect. We build the chatbot on your website or in WhatsApp and connect it to your knowledge sources.
4. Test with real questions. Not with a demo, but with the questions your customers actually asked.
5. Go live and review. After that we read the conversations back regularly and adjust the sources.

How long this takes depends on the number of questions, the state of your sources and the integrations. After the count in step 1, you know. How often we read conversations back afterwards is agreed per project.

Your Tech Club is certified for ISO 27001 and ISO 9001 by Kiwa. That certification covers our organisation and way of working. What happens to your chatbot's conversations, who may read them and how long they are kept, we agree per project.

What do you need to sort out for the EU AI Act and GDPR?

Since 2 August 2026, a chatbot has to let people know they are talking to AI. That is in Article 50 of the EU AI Act, and the notice has to come at the latest at the first interaction. Only when it is already obvious to an attentive user is it not required. On top of that, the GDPR simply applies, because a chatbot often processes personal data.

The rule is in Article 50(1) and 50(5) of Regulation (EU) 2024/1689 (EUR-Lex, 2024, accessed October 2026). The notice has to be clear and distinguishable. A line such as "You are chatting with our AI assistant" above the chat window is therefore no longer a nice extra.

For the GDPR I mainly look at three questions. Where do the conversations go? How long do you keep them? And do you have a data processing agreement with every party that reads along? What else you need to arrange is covered in EU AI Act and GDPR: what to sort out.

Where do you start if you want a chatbot?

Start with your inbox, not with a tool or a chatbot development agency. Count for one week which questions come in and how much time they take. Then you can see whether a FAQ page, a chatbot or an AI agent delivers the most. That order stops you from building something that looks great in a demo and then gathers dust.

Want to work that out with someone who does this every day? Then we do it in the AI Opportunity Audit. You get a report with the questions that cost the most time, the solution that fits them and also the places where you are better off building nothing. That report is yours, even if you carry on with another party afterwards. We run the audit in English too.

Why count first? Looking back on a retailer running several AI agents, I would have asked even earlier which automations were already running. That is why we now often start with an audit: opportunity first, then the technology.