Chatbots in customer service

What is a customer service chatbot?

A customer service chatbot is a dialogue system that automatically answers written customer inquiries and hands off issues it cannot resolve to human agents. To do this, the chatbot identifies the intent using fixed rules or artificial intelligence and retrieves the answer from a knowledge base or connected systems.

How does a customer service chatbot work?

A customer service chatbot receives a customer's message and assigns it to a specific intent, such as a contract question or an address change. A rule-based chatbot recognizes the intent based on keywords or buttons clicked by the customer. An AI chatbot, on the other hand, understands freely formulated sentences. Such chatbots are part of AI automation, which refers to the automation of customer interactions using artificial intelligence.

The chatbot then compiles the answer. It answers general questions using a knowledge base of approved content. For personal information, such as the status of a claim, it first identifies the customer and retrieves the data from the connected backend system.

If the chatbot cannot resolve the issue, it transfers the request to a human agent. The previous dialogue is passed along, so the customer does not have to repeat their request.

What types of chatbots are there?

In customer service, there are three types of chatbots: rule-based chatbots, AI chatbots, and hybrid chatbots. The key difference lies in how a chatbot identifies intents and finds answers.

Rule-based chatbots follow a fixed dialogue tree. The customer chooses from predefined answers, or the chatbot reacts to specific keywords. A special form of these are click-bots, which only offer buttons and do not allow for free-text input.

AI chatbots, often called intelligent chatbots, recognize the intent behind a freely formulated message. Older systems map these to trained intents and output a stored response. Generative AI chatbots based on large language models go a step further by formulating the answer themselves, drawing on approved company content.

Hybrid chatbots combine both approaches. Fixed dialogue paths secure steps that must follow a precise sequence, such as identification, while the AI handles free-form questions. Beyond these three types is Agentic AI: Such a system determines the solution path for an inquiry itself and executes it within connected systems.

Rule-based chatbot vs. AI chatbot: the differences

Both types answer customer inquiries automatically. The difference lies in how freely customers can express themselves and how much maintenance the chatbot requires.

Understanding: A rule-based chatbot only recognizes inputs that match its rules. An AI chatbot also understands typos, colloquial language, and phrasings that no one has previously defined.

Responses: A rule-based chatbot only outputs pre-defined text. A generative AI chatbot formulates its own responses and therefore requires a firm connection to verified sources.

Maintenance effort: With a rule-based chatbot, every new question must be added as a rule. An AI chatbot understands new phrasing without additional rules, but requires maintained knowledge content for new topics.

Control: The responses of a rule-based chatbot are completely predictable. With an AI chatbot, testing, approvals, and source binding ensure quality.

A rule-based chatbot is suitable for requests with a fixed process, such as booking an appointment. If many different questions are received in free-form language, an AI chatbot is the more appropriate choice.

Chatbot, Conversational AI, Voicebot, and Live Chat: The Differences

Three terms are frequently confused with the chatbot in customer service.

Conversational AI: Conversational AI is the AI technology used by systems to conduct dialogues in natural language, both in writing and over the phone. An AI chatbot uses this technology in the written channel, while a rule-based chatbot functions without it.

Voicebot: A voicebot conducts automated dialogues over the phone and uses speech recognition and speech synthesis to do so.

Live Chat: In a live chat, an employee responds in real time, whereas a chatbot responds automatically and transfers to a live chat when necessary.

Where are chatbots used in customer service?

In customer service, chatbots primarily handle frequent requests with clear answers. These include information on opening hours, responsibilities, or required documents. Additionally, they handle status inquiries, such as the status of a claim, an application, or a delivery.

Simple tasks can also be completed via chat. A customer reports a new address, requests a contribution certificate, or schedules a callback. The chatbot collects all the information necessary for processing.

For complex requests, the chatbot handles pre-qualification. It asks about the request, verifies identity, and transfers the case with all details to the responsible employee. Typical examples include insurance claims and contract questions, administrative questions regarding applications and appointments, and retail orders and returns.

What are the advantages and limitations of a chatbot in customer service?

A chatbot answers standard questions around the clock, including evenings and weekends. Because it handles many dialogues in parallel, it absorbs peak loads, such as after a storm with many damage reports. Your employees gain time for requests that require consultation or a decision. Transferred cases also arrive pre-qualified and can be processed more quickly.

Chatbots have their limits when it comes to requests requiring discretion or empathy, such as handling a complaint. Rule-based chatbots also fail when faced with questions outside their programmed paths. Generative AI chatbots can produce incorrect answers if they are not grounded in verified sources. If there is no clear handoff to a human agent, customers can get stuck in a loop.

What to consider when introducing a chatbot in customer service

Before launching a chatbot in customer service, several points should be clarified.

Suitable requests: A chatbot is worthwhile for frequent, well-structured requests. If you only receive a few chat inquiries, a live chat or a well-maintained FAQ section is often the more cost-effective solution.

Knowledge base: The quality of the answers depends on up-to-date, approved content. Maintaining this requires a dedicated person responsible within the department.

Handoff to employees: The chatbot should recognize when it has reached its limits and hand over the conversation along with all relevant information. Customers should also be able to request a human agent at any time.

System integration: For status inquiries and updates, the chatbot needs access to customer and backend systems, as well as a secure identification process.

Data protection and operations: Customer dialogues contain personal data. Therefore, it must be clarified where the chatbot and language model are hosted and who has access to the data.

Chatbots in customer service in practice

A typical starting point is handling requests that arrive repeatedly via contact forms or phone, such as questions about order status. It is crucial that everything the chatbot passes on reaches the responsible employee directly.

In Dynamic Workload, the chatbot is part of the Conversational AI within the AI Automation module. The Conversational AI conducts dialogues around the clock across all channels and in 120 languages. It is operated in the IPD Sovereign Cloud, exclusively in German data centers. Requests that it cannot resolve are handed over to employees with full context. Distribution is handled by Workload Management, which bundles requests from all channels into a central work queue and assigns them to the appropriate person.

How this technology works over the phone is demonstrated in the Tchibo Success Story. There, the AI voicebot "Voicy" handles request recognition as well as caller identification and authentication for 2,400 calls per day. Since its implementation, call times have been reduced by up to 30 seconds.

FAQ about chatbots in customer service

Häufig gestellte Fragen

A customer service chatbot usually runs as a chat window on a website or within a customer portal. It can also be integrated into apps and messaging platforms like WhatsApp. It is important that all channels access the same knowledge base. This ensures that customers receive the same information everywhere, regardless of where they ask.

A well-configured chatbot recognizes when it cannot confidently answer a question and offers to hand the conversation over to a human agent. Depending on the time of day and the organization, this can take place via live chat, a callback, or a support ticket. The previous dialogue is transferred so that the customer does not have to repeat themselves.

Language models like the one behind ChatGPT can form the foundation of a generative AI chatbot. However, without a connection to company-specific content, such a model answers based on general knowledge and may provide incorrect information. In customer service, it is therefore linked to an approved knowledge base, a process known as Retrieval Augmented Generation (RAG). Additionally, data protection and operational requirements must be addressed.

A chatbot handles recurring, standard inquiries, thereby relieving the burden on customer service staff. Inquiries that require consultation, discretion, or case-by-case decision-making remain with humans, as do all cases escalated by the chatbot. This is based on the "human-in-the-loop" principle: the AI provides support, and the human makes the decisions.