AI Agent ROI Calculator
Free training: Chatbot crash course
Whitepaper: The acceptance of chatbots
Artificial General Intelligence (AGI)
--> Go to BOTwiki
Definition and delimitation of AGI
Artificial General Intelligence (AGI) refers to a hypothetical artificial system that has the ability to understand and learn any intellectual task that a human being can understand and learn. It is assumed that AGI will have at least equal or superior capabilities to humans in all cognitive and intellectual areas. The development of AGI is a central research goal in the field of artificial intelligence, but has not yet been achieved.
In contrast, the artificial intelligence that is widely used today is referred to as artificial narrow intelligence (ANI) or weak AI. ANI systems are limited to specific tasks and areas of application and cannot perform in other domains. Examples of this include current chatbots, voice assistants, and image recognition systems. AGI, on the other hand, would be universally applicable and could transfer and apply knowledge and skills across different areas.
Properties and characteristics of an AGI
Artificial general intelligence would replicate the entire spectrum of human cognitive and intellectual abilities. These include learning ability, understanding natural language, the ability to reason and plan, problem-solving skills, and the ability to adapt to new situations. Such a system would possess "common sense" and be able to solve complex problems in environments and contexts for which it has not been explicitly trained. Open questions remain regarding the consciousness or sentience of AGI, which also touches on philosophical dimensions of the research.
Current status and research perspectives
Despite significant advances in the field of artificial intelligence, artificial general intelligence is still considered a distant dream. Current AI systems, including advanced language models, demonstrate remarkable performance in specific areas, but do not yet reach the level of AGI. It is believed that the development of AGI will require various AI concepts and technologies such as machine learning, deep learning, artificial neural networks, natural language processing (NLP) and computer vision. Researchers have different estimates of the exact time frame for the possible realization of AGI, with predictions ranging from a few years (interview with Demis Hassabis, CEO of Deepmind, and Sergey Brin, co-founder of Google) to several decades.
AGI in the context of conversational AI and AI agents
In the field of conversational AI and the development of AI agents at BOTfriends, specialized AI systems are used. These systems, such as chatbots and voicebots, are examples of artificial narrow intelligence, as they are trained for specific conversation flows and tasks. Future artificial general intelligence could fundamentally change the possibilities of conversational AI and AI agents. AGIs could be capable of conducting highly complex dialogues autonomously, applying knowledge flexibly, and adapting independently to new conversation requirements and topics without the need for extensive reconfiguration or retraining. The vision of AGI therefore represents a long-term horizon for the further development of intelligent automation solutions.
Frequently Asked Questions (FAQ)
Artificial General Intelligence (AGI) is characterized by the ability to solve any intellectual task at human level or better, without being limited to a specific domain. In contrast, conventional or "weak" AI (Artificial Narrow Intelligence) specializes in performing specific tasks in predefined domains. Current chatbots or image recognition systems fall into the category of weak AI, while AGI would be a universally applicable and adaptive system.
There is no general consensus on the exact point in time when fully functional artificial general intelligence (AGI) could become a reality. Expert estimates vary considerably, ranging from possible development within the next five to ten years to forecasts that estimate it will take another 20 years or more. Despite their advanced capabilities, current AI systems are not yet considered AGI.
For conversational AI systems, such as those developed by BOTfriends, artificial general intelligence (AGI) would have a transformative significance. Today's conversational AI is trained for specific use cases. AGI, on the other hand, could independently develop new conversation strategies, use knowledge across contexts, and adapt flexibly to unforeseen conversation situations, which would mean far greater autonomy and problem-solving ability in interaction. AGI thus represents a long-term goal in the development of highly intelligent conversation systems.
> Back to BOTwiki
Hybrid Human Chatbot
--> Go to BOTwiki
A hybrid human chatbot is a solution in which chatbot and human service representatives work together to handle customer communications. The chatbot first processes incoming inquiries automatically. If it cannot answer a question or the situation requires personal assistance, it seamlessly transfers the inquiry to a human agent: this is known as human handover. For the user, this switch takes place within the same chat window, without any media discontinuity. The human employees receive the complete conversation history and context, so that no information is lost. At the same time, the chatbot learns from the human responses and continuously expands its knowledge base.
Why is Hybrid Human Chatbot important?
The hybrid approach is particularly valuable for companies in Germany, as it combines efficiency with service quality. Around 80% of customer inquiries are recurring and can be answered by a chatbot in a matter of seconds—24/7, without any waiting times. At the same time, human employees remain available for complex issues, emotional situations, or high-quality consultations. This leads to increased customer satisfaction, as each inquiry is assigned to the appropriate agent. For service teams, this means less routine inquiries and more time for challenging cases. The advantage is particularly evident during peak periods such as the Christmas season: the chatbot acts as an intelligent firewall that filters and prioritizes inquiries.
Hybrid human chatbot in practice
A typical example of application is customer service in e-commerce: The chatbot automatically answers questions about shipping times, returns, or product availability. For customers with high shopping cart values or complex product consultations, the chatbot transfers them to a service representative.
BOTfriends enables such hybrid solutions through seamless integration with a wide variety of live chat systems, such as Zendesk or Userlike. The platform continuously analyzes unanswered questions and automatically suggests new topics, so that the chatbot constantly learns and improves.
Frequently Asked Questions (FAQ)
The human handover takes place seamlessly in the same chat window. The chatbot uses intent analysis, sentiment, or defined triggers to recognize when a human takeover is necessary. The agent receives the complete conversation history, customer history, and context-related information. BOTfriends offers this feature with integration into common live chat systems, eliminating the need for duplicate infrastructure.
Hybrid solutions combine the best of both worlds: 24/7 availability and scalability of AI with human empathy and problem-solving skills. This prevents frustration due to bot limitations, increases the first-contact resolution rate, and enables continuous learning. Companies can start with a smaller bot scope and expand it organically based on real user queries.
The hybrid approach is particularly suitable for customer service, e-commerce, human resources, IT support, and sales. This solution is ideal wherever a wide range of questions is expected or certain process steps require human decisions. BOTfriends supports companies in identifying suitable use cases and offers a testing option prior to full automation using the Wizard of Oz method.
--> Back to BOTwiki
Prebuilts
--> Go to BOTwiki
Prebuilts, or ready-made knowledge bases, are a collection of content for specific chatbot use cases. Chatbots live on content that is taught to them. To develop a chatbot, it is important to enrich it with content, i.e. questions and answers. The more content, the broader its knowledge and thus the answers it can give. Prebuilts help with the development and can be used as "boosters". If you want to build a chatbot in the area of IT support, for example, it is very helpful to be able to fall back on such prefabricated knowledge bases so that a basic stock of content is already available depending on the use case.
How are Prebuilts structured?
Prebuilts consist of question options, entities and answers. To increase the speech recognition of a chatbot, it is important to store many different question options for an intent. Furthermore, entities and answers belong to each intention. Prebuilt exemplary answers exist for prebuilts, which the customer can adopt 1:1, but of course also adapt to the company, character and tonalityof the chatbot.
How can Prebuilts be used?
Prebuilts are usually easy to integrate. In most cases, the prebuilts are integrated via the NLP service used. Here, there is often a one-click integration with which you can quickly and easily import the required content. Another possibility to work with prebuilts is the integration within Intent Management Platforms (IMP). These platforms are there to make content-specific changes quickly and easily.
Which Prebuilts are already used sensibly?
There are already many available sample prebuilts that can be integrated into a chatbot. The Smalltalk Prebuilt is certainly one of the most frequently used prebuilts. This also makes a lot of sense, because if you look at a typical communication of a user with a chatbot, especially at the beginning some smalltalk questions like "How are you?", "How old are you?" are asked. However, there are also other pre-builts, such as weather, news or restaurant bookings. Here you have to pay attention to the languages in which these prebuilts are available. All in all, prebuilts can provide start-up help in some areas and support the development of a chatbot.
> Back to BOTwiki
Chatbot
--> Go to BOTwiki
A chatbot is a computer program designed to conduct conversations with people via digital channels and automatically answer inquiries. The term is a combination of the English word “chat” for conversation and “bot,” short for “robot.” Chatbots are integrated into websites, messaging services, or other digital interfaces and often process user input using artificial intelligence.
Differences from conventional bots
In computer science, the term "bot" generally refers to a program that runs autonomously, performing its tasks without human intervention and requiring only a single initial setup. Bots can serve various purposes, including malicious ones, such as in botnets, which enable coordinated denial-of-service attacks.
A chatbot is a specialized form of this concept. It is explicitly designed for communication with humans and is geared toward processing conversations in natural language and providing meaningful responses. The key difference lies in its ability to interact: While general-purpose bots perform tasks in the background, a chatbot is designed for direct interaction with users.
The Role of Artificial Intelligence
The processing of natural language input is only possible through the use of artificial intelligence. Human language does not follow rigid rules and cannot be processed directly using traditional programming logic. Instead, Natural Language Processing , or NLP for short. In this process, a user’s input is interpreted by so-called large language models to identify the underlying intent.
Once an intent is recognized, it is converted into an action. This action can be a direct response to a question or the execution of a process, such as retrieving information from a connected system. For a chatbot to provide meaningful responses, the NLP component must be backed by a structured knowledge base containing answers and defined actions.
Applications and Channel Integration
Chatbots are used across a wide range of text-based channels. In customer service, they are often set up as web chats on a company’s website or integrated into social media platforms such as Facebook Messenger or Instagram. They can also be integrated into HR workflows via Slack or Microsoft Teams.
Various platforms are available for developing and operating chatbots. Providers such as Google (with Dialogflow), IBM (with Watson), and Microsoft (with LUIS) offer their systems as publicly accessible web services. These platforms provide access to sophisticated machine learning systemswithout the need to build your own infrastructure.
BOTfriends offers a high-performance enterprise platform that allows you to build intelligent chatbots in-house and manage them centrally across all channels. Thanks to its intuitive no-code architecture, you can automate complex customer interactions all the way to the backend without relying on expensive external integrators.
A chatbot processes text-based inputs and provides text-based responses. A voicebot, on the other hand, is designed for spoken language and also uses speech recognition and speech synthesis technologies. In practice, both types can be based on the same conversational logic and differ only in their input and output modalities. BOTfriends X supports both variants within a single platform.
Not necessarily. There are rule-based chatbots that rely on fixed decision trees and predefined response paths without using AI components. These are suitable for simple, clearly structured use cases. However, as soon as natural language inputs in variable forms need to be processed, the use of NLP—and thus AI methods—is necessary to ensure reliable recognition of user intent.
Chatbots are used in a variety of fields, such as customer service to answer frequently asked questions, internal IT support, recruitment, and e-commerce. They are deployed in situations where there is a high volume of similar inquiries and automated processing is beneficial. They are integrated through channels such as web chat, WhatsApp, Microsoft Teams, or other messaging platforms.
> Back to BOTwiki
Conversational office
--> Go to BOTwiki
A conversational office, also known as a conversational workplace, refers to a business environment in which a large portion of internal communication and service processes is handled through conversational systems. Instead of navigating through intranet menus, self-service portals, or ticket forms, employees speak or write directly to an AI agentwho understands their requests, retrieves data from the appropriate backend systems, and initiates the necessary actions.
This creates a new communication standard for companies that brings together HR, IT support, and facility management on a unified conversational platform.
What Makes a Conversational Office
A conversational office is more than just a single FAQ bot on the intranet. It combines multiple AI agents, each of which covers a specific area of expertise—such as human resources, IT services, travel management, or office logistics. Through multi-agent orchestration, these agents access shared identities, permissions, and knowledge sources, and seamlessly forward inquiries to the appropriate specialist agent.
Technically, a Conversational Office is based on a unified conversational AI platform that integrates voice, chat, and email under a single logic. Knowledge AI ensures that internal policies, manuals, and tool documentation are accessible without employees needing to know which system contains the answer.
Typical areas of application within a company
The applications range from recruiting to office management. A Conversational Office brings these use cases together into a single, unified conversational experience instead of spreading them across ten different tools.
- HR and Recruiting: Job searches, application status, vacation requests, pay stubs, and onboarding steps are handled through the dialog.
- IT Support: Password resets, device orders, software approvals, and trouble tickets are handled via AI workflows and escalated as needed.
- Office services: Room reservations, cafeteria menus, shuttle schedules, visitor registration, and taxi orders can all be handled via self-service.
- Knowledge Access: Employees ask about policies, processes, or contract terms and receive answers backed by reliable sources.
Implications for Voice and Chat
In the Conversational Office, the voice channel is playing an increasingly important role. An internal service hotline that used to be handled by a traditional IVR system can now be managed by a phonebot that immediately understands the issue and directs the caller to the right place without a menu of button presses. Traditional IVR is a body without a brain. AI-native voice with multi-agent orchestration, on the other hand, combines natural language understanding with actual task processing. An employee who needs to reset their password while on the go can do so over the phone instead of in a browser.
In a chat, such as via Microsoft Teams, Slack, or an internal web widget. Does the voice experience complement intent-based dialogues featuring attachments, buttons, and structured responses. Email inquiries sent to service inboxes can also be automatically classified and answered by the same AI agent, ensuring a consistent knowledge base across all channels.
Conversational Office in Multi-Agent Setups
A productive Conversational Office rarely consists of a single all-around agent. Hybrid intelligence—combining rule-based process logic and generative models—forms the brain, while individual specialized agents act as the body for HR, IT, or facilities. An orchestration layer determines which agent handles a request, when a human employee takes over, and what information flows between the systems. This allows the Conversational Office to scale with the company’s structures, rather than managing each department in an isolated tool.
> Back to BOTwiki
Conversational testing
--> Go to BOTwiki
Conversational testing refers to the systematic testing of the processes defined in the conversation flow for an AI agent for naturalness and comprehensibility before they go into production.
The goal is to identify early on whether the wording sounds natural, whether the dialogue achieves its objective, and whether the desired tone is maintained. This process is a central component of conversational design and supplements automated testing methods with human evaluation. In this way, conversational testing combines traditional quality assurance with the hybrid intelligence requirements of modern conversational AI.
What Conversational Testing Reveals
Conversational testing reveals areas where the AI agent's conversational capabilities still need improvement. These include:
- Convoluted or tediously long sentences that don't work in voice chat.
- Missing or unclear follow-up questions when the intent is ambiguous.
- Shifts in tone between formal and informal passages.
- Gaps in the conversation flow where users cannot identify a logical next step.
- Answers that are factually correct but miss the point.
Implications for Voice and Chat
In the voice channel, for example, with a voicebot in hotline triage, conversational testing is particularly valuable. Spoken language does not tolerate convoluted constructions, and users expect short, clear responses.
In the context of chat and email, the focus shifts to readability, tone, and striking the right balance between precision and empathy. Here, too, testing reveals whether responses are perceived as helpful or whether users need to ask follow-up questions to clarify the issue.
Conversational Testing in Multi-Agent Setups
In complex scenarios, multiple specialized AI agents work together—for example, for authentication, case handling, and escalation. Conversational testing becomes particularly relevant here at the handoff stage, because gaps in communication between agents can quickly lead to repetitions or lost contextual information. In conjunction with Knowledge AI and defined AI workflows, this approach helps identify process boundaries and clearly delineate the areas of responsibility for each agent.
For effective implementation, an iterative approach is recommended: the results of testing are incorporated into revised training phrases, adjusted fallback paths, and refined workflow steps. This leads to continuous improvement, making conversational AI significantly more robust over time.
Frequently Asked Questions (FAQ)
It makes sense to use this approach once a conversation flow has been roughly established and the key responses have been formulated. In practice, testing takes place between the conversational copywriting phase and technical implementation in order to identify weaknesses early on. However, it can also be repeated later for new use cases or during major revisions to the dialogue.
Automated tests primarily assess the NLU model’s recognition performance and the technical stability of workflows. Conversational testing supplements this by incorporating human evaluation of tone, conversational flow, and perceived helpfulness. Both approaches are complementary and should be used together in professional conversational AI development.
> Back to BOTwiki
Insult Rate
--> Go to BOTwiki
The insult rate is a metric from chatbot or voicebot analytics that indicates how many conversations users insult or abuse the bot in. It is calculated as the ratio of conversations with insults to the total number of interactions. Insults often occur when the AI agent does not provide a suitable response, causes misunderstandings, or reaches its technical limits. Sometimes insults are made for no reason, for example by so-called trolls who take advantage of the anonymity of the internet. The insult rate is therefore an indicator of frustration, unfulfilled expectations, or poor user experience.
Why is the insult rate important?
The insult rate is a key metric for evaluating efficiency and user acceptance. A high rate often signals technical or content-related deficiencies: incorrect responses, poor issue recognition, or inadequate dialogue management. However, it can also depend on the target group or use case. In sensitive areas or among younger users, the inhibition threshold for insults often increases. The insult rate is valuable for companies because it identifies specific areas for optimization: Which dialogs cause frustration? Where are answers missing? Systematic evaluation of this KPI helps to continuously improve the performance of chatbots or voicebots and increase customer satisfaction.
Insult rate in practice
In practice, the insult rate is used to identify specific weaknesses in chatbot dialogues. Example: A customer service bot uses offensive language in 15% of all conversations. Analysis shows that users react with frustration, especially when asked about delivery times. The Chatbot does not provide clear answers here. After optimizing the intents and responses, the rate drops to 5%.
BOTfriends relies on data-driven chatbot optimization: dialogues are continuously improved by evaluating the insult rate and specifically training the AI models. This creates a positive user experience that minimizes insults and sustainably increases acceptance of the chatbot.
Frequently Asked Questions (FAQ)
The insult rate is calculated as a percentage: the number of conversations containing insults divided by the total number of conversations, multiplied by 100. Example: 3 insults in 30 chats result in an insult rate of 10%. Detection is usually carried out using keyword analysis or natural language processing (NLP).
A high insult rate indicates problems: poor response quality, missing intents, technical errors, or a frustrating user experience. It is a warning sign that the chatbot or voicebot needs to be revised. BOTfriends systematically analyzes such cases in order to implement targeted improvements.
Yes, the insult rate depends heavily on the target group and use case. In sensitive areas such as health or finance, it is often lower, while it is higher among younger target groups or in informal contexts. The time of day and anonymity also have a significant influence on user behavior.
> Back to BOTwiki
Quick Reply / Chips / Chips
--> Go to BOTwiki
Quick Replies are a form of interaction in messaging platforms and appear as buttons under a message. Unlike cards, Quick Replies disappear after they have been pressed. Quick replies allow users to click through content in the chatbot instead of typing their request in the input field. If, for example, the chatbot offers a button with the name "Help", it is no different in the end than if the user types in "Help" and sends it. The communication is thus accelerated and the users get an orientation about the topics that the chatbot offers.
Depending on the output channel or messaging platform, different names are used for Quick Replies.
| Messaging platform | Designation |
| Facebook Messenger | Quick Reply, Quick Reply |
| Google Assistant | Suggestion Chips |
| Microsoft Bot Framework | Suggested Actions |
| slack | Slack interactive buttons |
| Skype | HeroCard |
| telegram | Keyboard Buttons |
| viber | keyboards |
Here is an example picture for Quick Replies in Facebook Messenger:
Application in Dialogflow
If you create or edit intents in Dialogflow, you will find several options for replying as a chatbot under the heading Response. Among other things, you have the option of selecting the Quick Reply format and naming the buttons in the form of a list.
A Quick Reply Response is displayed in the messaging platforms as a predefined User Response, which is sent back to Dialogflow by clicking on it.
> Back to BOTwiki
AI Agent Training
--> Go to BOTwiki
AI agent training refers to the continuous improvement of an intelligent digital assistant’s response quality, contextual understanding, and behavior. In the age of generative AI, this involves less manual assignment of intents and focuses primarily on optimizing instruction prompts, refining the knowledge base, and fine-tuning the AI persona.
The goal is for an AI agent to not only understand user queries but also to respond in the right tone, with accurate facts, and within defined parameters. Training is an ongoing process that ensures the AI’s generative freedom remains aligned with business requirements and the reality of user interactions.
What Happens During AI Agent Training
At its core, the goal is to refine the agent’s interaction logic and knowledge base. Instead of rigid recognition patterns, the agent’s “brain” is trained through iterative prompt engineering and data maintenance.
A typical training cycle today includes:
-
Log Analysis & Evaluation: Analysis of dialogues for hallucinations, deviations in tone of voice, or gaps in knowledge.
-
Prompt iterations: Customizing the instructions in the instructions prompt and the AI agent persona to control the agent's behavior and communication style.
-
Knowledge refinement: Optimizing knowledge sources (documents, FAQs) so that Retrieval Augmented Generation (RAG) provides more accurate information.
-
Guardrail tuning: Refining safety filters to ensure that the agent does not make any unwanted or false promises.
When and how often to exercise
Training begins in the design phase with the creation of the AI persona and the dialogue logic. However, it is only after the system goes live that we see how users actually interact with the generative AI. Since LLMs (Large Language Models) can respond to inputs in unpredictable ways, close monitoring is essential.
-
Initial training: Developing the persona, defining areas of responsibility, and connecting the first knowledge sources.
-
Pilot phase: Test runs using a "human-in-the-loop" approach to validate the quality of the generated responses under real-world conditions.
-
Continuous optimization: Regular (weekly) analysis of user feedback and updates to the knowledge base.
-
On-the-spot training: Immediate updates to training materials in response to new company policies or product changes.
Implications for Voice and Chat
Training is particularly critical in the voice channel (e.g., AI hotline). Here, AI agents must learn to handle natural speech flow, interruptions, and acoustic misunderstandings. The training focuses on instructing the agent to grasp the essence of the request even with inaccurate speech-to-text input (dialects, background noise) and to guide the user to the desired outcome through agentic dialogues (active follow-up questions).
In chat and email contexts, the focus is on information density and formal accuracy. The training ensures that agents can accurately summarize complex documents and provide precise written instructions without overwhelming users with overly long blocks of text.
AI Agent Training in Multi-Agent Systems
In modern architectures with multi-agent orchestration, training becomes modular. Instead of training a monolithic system, one trains specialized agents for specific tasks (e.g., a “technical expert” and a “contract assistant”). Each agent receives specific training for its domain:
-
One is trained to make precise API calls (tools).
-
The other is optimized for empathetic problem-solving (Persona). An overarching orchestrator manages the context. In this context, hybrid intelligence means that the training continuously improves both factual knowledge (Knowledge AI) and procedural intelligence (Agentic Workflows).
Frequently Asked Questions (FAQ)
It involves the continuous fine-tuning of prompts and knowledge sources (RAG) to maximize the quality and accuracy of an AI agent’s responses.
AI models are versatile, but they don’t know your specific business rules, products, or brand voice. Training “teaches” the AI to behave exactly as your brand requires.
To some extent. While Knowledge AI (RAG) automatically provides the agent with up-to-date information, manually training the instructions remains important for controlling how that information is conveyed (e.g., in a friendly, factual, or sales-oriented manner).
For voice systems, training focuses on robustness against speech errors and real-time dialogue control. The agent must be trained to interpret pauses correctly and actively guide callers through processes, rather than simply waiting passively for text input.
> Back to BOTwiki
edge case
--> Go to BOTwiki
Edge cases are outcomes of a conversation that are not expected , that rarely occur and therefore represent exceptions in the conversation . These edge cases are created, among other things, within the Conversational Map or the Conversation Flows. The counterpart to edge cases is the Happy Paths are. Here, the expected outcomes of a conversation are described, which are most frequently taken by the user.
Example for the Happy Path on the Use Case "Order Pizza
This example shows what a "happy" conversation between user and chatbot can look like when ordering a pizza. All information that the user provides to the chatbot can be processed and no misunderstandings arise.
Example for the Edge Case at the Use Case "Order Pizza
This example shows that there can also be great potential for error if the user sends answers that the chatbot cannot process. The graphic below shows that the user enters an address that is outside of Germany. In this case, no delivery can take place. Such edge cases should be considered in advance and mapped in the conversational map. You should ask yourself how to deal with the user in such situations. For example, you can inform the user that you only deliver within Germany or you can give them the option to enter the address again in case of a misunderstanding.
> Back to BOTwiki