Consulting & Innovation
Solutions & Products
Infrastructure & Operations
Solutions & Products
Titelbild_Blogartikel_Die Arten von Chatbots

The different types of chatbots

The different types of chatbots

IKEA does it, Amazon too, and Google anyway. Chatbots are becoming increasingly popular in B2C communication. Thanks to cloud technologies, the bots' understanding of language is growing steadily and enabling more and more genuine interaction between humans and machines.

In recent years, great progress has been made in text understanding by machines. This is enabling increasingly rich human-machine interactions in a way that humans are accustomed to: Through speech. One application of this technology that is relevant to the broad mass of customers is chatbots. These are used in messengers such as WhatsApp, in apps or on corporate websites for a wide variety of purposes. Here, however, chatbots differ greatly in their interaction possibilities. In this article, I explain the different types of chatbots.

Chatbots can be distinguished from one another in the dimension of interaction complexity, among other things. This will be looked at in more detail below.

Bild01_Blogartikel_Die Arten von Chatbots
Einteilung von Chatbots nach Interaktionskomplexität (Quelle: Arvato CRM - Portfolio & Solution Design)

Classification of chatbots according to interaction complexity

Source: Arvato CRM - Portfolio & Solution Design

QnA Bots

These bots are an easy way to provide extensive information services such as FAQ areas or libraries of user manuals for the customer via chat. Since customers still like to use keywords in such use cases in particular, several language processing variants come into question. For complex issues, where a high relevance rate in the answers is critical for success, so-called knowledge engineers train and configure a matching system manually. In doing so, they are able to respond particularly well to the specifics of the domain in question. If larger amounts of information with a simple character are to be processed, on the other hand, approaches based on machine learning can be applied. To do this, information is uploaded to a system such as Microsoft Azure QnAMaker. The system trains itself based on the data entered and can then be refined with additional question options. Once the bot is deployed, it accepts user queries and displays information from its knowledge base that is as appropriate as possible. Based on customer feedback, the bot learns whether its suggestions were useful or not.

For companies that receive a lot of customer inquiries, FAQ bots offer a modern and attractive way to have a high proportion of their customer inquiries answered automatically in self-service. This reduces customer service costs and at the same time provides valuable information for further product development.


Scriptbots offer the user a higher degree of interaction. Typically, they are used for highly standardized processes. For example, the user can be asked for the required information instead of entering it into a web form or announcing it to a call center agent.

Another use case is the configuration of products or services using a questionnaire-like process. The FinTech start-up "vaamo", for example, uses a scriptbot to determine the risk preference of its customers. With these bots, an understanding of the customer's language is not absolutely necessary, as the answer options are often limited to a few predefined options. Scriptbots relieve customer support of highly standardized business processes and in some cases can carry them out much faster than would be possible using other support channels.

 Natural-Language-Understanding (NLU) Chatbots

Natural-Language-Understanding (NLU) Chatbots, i.e. those with an understanding of natural language, offer even broader possibilities for interaction. They can analyze the texts written by users and thus recognize their concerns, so-called entities (names, times, customer numbers, etc.) or even mood. This allows the chatbot to be more flexible in responding to users and also to design processes accordingly.

For example, if a user writes: "I am now looking for a train connection from Munich to Vienna", the chatbot deduces that the user wants to book a train with departure point Munich, arrival point Vienna and the whole thing at the current time. Now the chatbot checks which information is still missing to complete the booking (e.g. the booking class) and asks for it afterwards. This procedure is also called "slot filling". A simple scriptbot, on the other hand, asks for each piece of information individually and one after the other - regardless of whether it was already mentioned in the user's initial message.

In addition, the mood of the customer can be detected dynamically. This offers the opportunity to treat each customer differently and increase their satisfaction. An upset customer is specifically passed on to a human call center agent, while the chatbot attends to the relaxed prospect looking for offers. This allows the company's resources to be deployed in a more targeted manner, reducing the churn rate and increasing customer satisfaction.

Virtual Agents

The most complex chatbots are the virtual agents. They extend the range of functions of the NLU chatbots with very flexible dialog structures and a perfect context memory. Queries such as "What is the fastest way home?" can be understood by these chatbots. For this purpose, the user's place of residence, the GPS of his device and connected information systems (e.g. GoogleMaps / local public transport app / MyTaxi app) are accessed. The information and accesses provided by the user as well as the connected systems are therefore particularly crucial for the intelligence of the chatbot.

Virtual agents can also leave the current dialog flow if the user suddenly expresses a different request. If the chatbot wants to inquire about the desired class of train, but the user just remembers that he needs to reserve a table at his favorite restaurant, the virtual agent would be able to master this context change and then return to the previous request. Full-fledged virtual agents do not yet exist, but it is only a matter of time before the first representatives of this type appear on the market.

Unused potential

It is clear that the more complex the interaction possibilities of chatbots, the greater the development and training effort. The potential of the technology is far from exhausted. It is becoming clear that companies will benefit greatly from chatbots in the design of their customer communications. At the same time, voice assistants such as Alexa and Google Assistant are gaining in importance. These are primarily controlled via spoken language and attempt to establish themselves as an overarching meta-assistant instead of offering capabilities in a specific area. Even if they are currently used for rather simple use cases (weather, sports results), the market is developing rapidly and a precise evaluation of the possible applications can be very worthwhile for companies.

AI in Practice

Be inspired by our application examples for AI. Discover our AI cases from practice.

Arvato Systems AI Services

We bring artificial intelligence to your business. Discover our services and solutions around AI here.

Deep Dive in Bot technologies

How exactly does a chatbot work? We explore the technology behind natural language understanding (NLU) chatbots.

Written by

Sebastian Stephan
Experte für Cloud Architekturen