*Thisarticle is based on a YouTube video by Sophie Hundertmark, an expert in the use of artificial intelligence with a focus on chatbots and strategic AI applications in companies and public institutions. Sophie is a researcher and lecturer at the Lucerne University of Applied Sciences and Arts and is doing her doctorate in Conversational AI at the University of Fribourg. The blog text was created using a custom GPT model that was trained on Sophie’s video content, language style and expertise. The result is well-founded, up-to-date articles based on Sophie Hundertmark‘s own expertise.
You can find the link to the video at the end of this article.
ChatGPT, Gemini, Claude, and other AI assistants are increasingly changing the way we search for information online. Instead of visiting a company’s website and searching for the relevant information there, many users ask their question directly to an AI.
This raises a perfectly valid question for companies: Do we even need our own chatbot anymore?
That’s exactly what I talked about in a new episode of “Sophie’s Next AI Talk” with Robert Weber, one of the founders of MoinAI. Robert has been working with chatbots for over ten years—from the first Facebook Messenger bots to today’s AI agents, which can be deeply integrated into business processes.
The short answer to our initial question is: Yes, business chatbots can still be relevant. However, we need to rethink what tasks they take on and what added value they offer compared to general AI assistants.
The customer journey is increasingly starting before the company’s website
For a long time, the company website was one of the most important digital starting points on a customer journey. Users searched for a company on Google, opened its website, and looked up information there about products, services, or support options.
Generative AI is changing this process.
You can already ask a question today using ChatGPT, Gemini, Claude, or another AI assistant. In the best-case scenario, the user will receive a suitable answer right there. The company’s website—and thus the chatbot integrated into it—may not be visited until much later, if at all.
This brings two questions to the forefront:
How can I make my company visible in AI systems?
And at the same time:
What additional value can my company’s chatbot offer if a user does visit my website?
It’s precisely this second point that makes things particularly exciting.
Simple FAQ chatbots are coming under increasing pressure
If a chatbot merely answers simple questions, its value is limited today.
In the interview, Robert describes, for example, traditional FAQ chatbots that use fixed decision trees and sometimes even provide outdated answers. At the same time, modern large language models provide users with powerful systems that are already very capable of answering many typical first-level questions.
Companies should therefore not just ask:
Do we really need another chatbot?
The more meaningful question is:
What kind of chatbot do we need?
After all, as soon as a company’s chatbot can do more than simply relay publicly available information, its role changes.
Three Stages in the Development of Corporate Chatbots
In the interview, Robert divides chatbots into three different stages of development. I find this classification particularly helpful because it gives companies a good sense of where the actual added value of their own chatbot might lie.
1. FAQ Chatbots
The first level consists of traditional FAQ chatbots. They answer frequently asked questions and, in some cases, still use relatively rigid decision trees.
Bots like these can generally be helpful. At the same time, they face particularly stiff competition from ChatGPT and other generative AI systems. After all, if a user can get the same answer more quickly and conveniently through their preferred AI assistant, there’s little reason to use a separate company chatbot for that.
2. Chatbots That Read
The second stage takes it one step further. These chatbots can access a company’s information or systems and retrieve up-to-date data.
One example from our conversation is flight data. If a traveler wants to know whether their flight will depart on time or if the departure time has changed, they need up-to-date and reliable information.
This is exactly where a corporate chatbot can add value: It not only knows general information, but also accesses company-specific and, where applicable, up-to-date data.
3. Chatbots That Read and Write
Things get particularly interesting at the third level.
Here, the chatbot not only retrieves information from other systems, but can also perform actions within defined processes.
For example, a customer of an energy provider could submit their meter reading directly through the dialog box or modify certain details on their bill. Instead of clicking through various screens and forms, the chat itself becomes the user interface.
That is precisely where the key difference lies compared to a simple question-and-answer system: The chatbot doesn’t just answer questions—it actually helps users complete tasks.
From a General-Purpose Chatbot to a Specialized AI Agent
An interesting real-world example from our conversation is Nuremberg Airport.
Specialized AI agents are used there for various tasks. These include, for example, specialists in baggage handling, parking, or check-in. Each agent has a clearly defined scope of responsibilities and works with a defined knowledge base that is maintained and updated by the airport.
This highlights an important trend: Companies don’t necessarily have to build a single AI assistant that’s expected to do everything right away.
Instead, various specialized agents may emerge.
An agent is particularly knowledgeable about luggage issues. Another one helps with parking. Another one answers questions about check-in.
This allows AI to be integrated into existing processes in a step-by-step manner.
Why Companies Don’t Need the Perfect AI Agent Right Away
Especially with AI projects, there’s a tendency to try to cover as many use cases as possible right from the start. However, this can make projects unnecessarily complex.
In the podcast, Robert therefore recommends an iterative approach: building one AI agent at a time and gradually expanding the system.
In my view, this is an important point for companies.
You don’t have to automate all your processes right away. Instead, start with a clearly defined use case that delivers real added value for your customers or employees.
If this works, you can add the next step.
Over time, this leads to the creation of a system of specialized AI agents, rather than launching a massive AI project that is expected to meet all requirements right from the start.
Good data remains the foundation
If you want to develop a powerful AI agent, there’s one topic you can’t ignore: your own knowledge base.
In the interview, Robert describes this step, in essence, as the “homework” that companies must complete first. The information should be well-maintained, structured, and available in a machine-readable format.
Structured databases, tables, or JSON files, for example, can be ideal. PDFs can also serve as a source of knowledge, but they should contain text that is as machine-readable as possible.
The first step, therefore, is often to consolidate different sources and build a knowledge base that is as centralized as possible.
After that, you might ask yourselves:
Which actions would we like to automate? Which processes are actually suitable for this? Where do we still need people? And what interfaces are already available?
APIs, webhooks, and MCP connect the chatbot to enterprise systems
If a chatbot is supposed to not only answer questions but also perform actions, it needs access to other systems.
This is where APIs or webhooks, for example, come into play. This allows applications to exchange information with one another or trigger actions.
The Model Context Protocol (MCP) also plays a role in this discussion. Simply put, MCP provides a standardized way to connect AI applications to external data sources, tools, and systems.
This technical aspect becomes particularly relevant when a chatbot is to be transformed into a true AI agent.
After all, an agent doesn’t just have to know what to do. It also requires the appropriate interfaces to be able to perform the desired action within an enterprise system.
When the chatbot learns from conversations what’s still missing
Another interesting concept that came up in our conversation is what MoinAI calls “Dreaming.”
This involves analyzing past user conversations to identify potential blind spots. For example, the system analyzes situations in which a chatbot was unable to provide a good response or in which users were dissatisfied with a response.
Such information is particularly valuable to companies.
That’s because chatbot conversations reveal very directly what customers actually know or want to do.
Instead of basing the further development of a bot solely on internal assumptions, real user interactions can provide insights into what information is missing and what additional features might be useful.
This makes the chatbot a source of customer insights as well.
Control and data sovereignty are becoming key differentiators
In addition to features and integrations, another factor plays an important role: control over the AI being used.
Data protection, the GDPR, and regulatory requirements are key issues, particularly for European companies. In the interview, Robert therefore emphasizes the importance of European AI providers and a European AI landscape.
For enterprise chatbots, this means that it’s not just about using the most powerful generative AI possible.
Companies must also consider what data is being processed, where it is being processed, what information a system is permitted to access, and to what extent the system’s responses and actions remain controllable.
These questions become even more important as AI agents are integrated more deeply into corporate systems.
The future doesn’t necessarily belong to the chatbot—but to the agent
So, do companies still need chatbots?
I think the answer depends a lot on what we mean by a chatbot.
A simple FAQ bot that merely provides publicly available information will find it increasingly difficult to succeed. ChatGPT, Gemini, Claude, and other systems can already easily answer many of these questions.
It gets more interesting when a business chatbot can do things that a general-purpose AI assistant can’t easily accomplish.
He is familiar with the latest internal company information. He accesses specific data in a controlled manner. He understands the context of each company. And above all: It can perform actions.
This shifts the discussion from traditional chatbots to AI agents.
For companies, this doesn’t mean they now have to rush to develop a massive AI agent as quickly as possible. It makes much more sense to start with a specific use case, clean up your own database, and then gradually integrate additional processes.
After all, the key question for the coming years may no longer be: “Do we need a chatbot?”
Rather: “What tasks can our AI agents actually take on for customers and employees?” Teach them!
Any further questions?
Do you have any questions? I am happy to support you, act as a sparring partner and answer your questions. I am always happy to receive your messages, preferably by WhatsApp message or e-mail.
