Create AI agents and general information

Learn how AI agents are optimised, created and defined for use cases

1. AI Agents – General information
2. Creating AI agents (two ways)
3. Best and worst use cases for AI agents
4. AI agents with follow-up AI actions
5. AI agent settings

AI Agents are specialists in their field. Based on generative AI (LLM), they respond individually and dynamically to incoming user queries. In contrast to the answers provided by intents, the AI agents' answers are not generic and repetitive, but react dynamically and individually to a user query. 

While the answer of a intent is based on the previously created answer content, selected resources form the knowledge of the AI agents.
The AI agent is activated via the 3-dot menu in the answer editor. Clicking on the Create agent button opens the option to change the intent.

Creating AI agents (two ways)

AI Agents can be created in two ways: indirectly from editorial topics and directly via the knowledge base. Both methods are described below.

1. Indirectly (from editorial topics)

   The Intents section contains a complete list of all editorial topics (and AI agents) created for the chatbot. Clicking on the pencil icon behind a topic opens the topic's edit page. The following describes how to convert the "Angebot" topic into an AI agent.

   
   In the editorial topic editor, the conversion is carried out via the 3-dot menu. The conversion is carried out with a click on the Create Agent button.

   
   Click Save to complete the change.
   
   

2. Directly (via the Knowledge Base)

   1. Add AI agent: AI Agents can be added behind the Knowledge Base menu item by clicking on the plus symbol in the top right-hand corner.
   
   
   2. AI Agent description: An input mask then opens in which the AI agent is described and defined. The purpose of the AI agent is defined in the first field. Clicking on Generate triggers the second step, in which the input is saved and validated.
   
   
   3. Comparison: The chatbot checks whether an existing AI agent already covers this topic. If not, clicking on Proceed at the bottom opens the next input field.
   
   
   4. AI Agent review: Here, the AI agent name and application description can be edited again. Once all changes have been made, the AI agent is reviewed by clicking Proceed.
   
   
   5. Sample questions: In the last step, 5 sample queries are listed, which the AI agent uses as a guide. These can be customised. Finally, the AI agent is created by clicking Create Agent.
   
   

Topic or agent not recognised

• If a different topic is recognised, this means that the question intention is already covered. There is no need for a new agent. If necessary, new sources may need to be added to the existing agent.
• If ‘unknown’ is displayed, the chatbot's use case is not sufficiently well defined. This is defined in the Bot settings -> Channel management -> Pencil icon -> Use Case.

AI Agents: Best and worst use-cases

AI Agents are primarily used to reduce the editorial effort involved in creating responses. They are therefore generally suitable in areas where a high editorial effort meets a low query volume (so-called long-tail queries). For intents with a higher volume of queries, an evaluation of the response documentation is necessary beforehand. If the documentation is good, the AI agents are also suitable here. 
The AI agents are also suitable where the answers are very repetitive and good documentation exists at the same time. 
However, the AI agents are unsuitable where the answers are dynamic and based on complex documentation, e.g. questions about individual product key figures. 
AI Agents are also unsuitable if the answer contains sensitive information, e.g. dosages of dietary supplements or information on allergens. 
Finally, an AI agent is not recommended if only a small amount of editorial work is required to create an answer.

AI agents with follow-up AI actions

The follow-up action allows you to specifically play another topic or form after an AI agent response. This enables you to continue the conversation in a meaningful way. Users are guided through the processes in a targeted manner. The following describes how to set up a follow-up AI action step by step.

Setting up a follow-up AI action

Under the Knowledge Base menu item, the RAG Knowledge Base view opens via the RAG button. A new input mask opens via Add Action in the Actions & Follow-Ups card.

In the Follow-up tab, select the topic to which users should be redirected to. Also, enter the yes-no-question for the follow-up message here. Click Save to create the redirection.

Add instruction in AI agent

In order for the follow-up action to actually be executed, an instruction must be created that instructs the AI agent to perform the follow-up action. How to create instructions is described in this article.

Example: Live chat offer

Two configuration examples are explained below – one without and one with a placeholder in the query.

1. Without placeholders in the query
   Scenario: An AI agent for cost questions should automatically offer live chat advice when a price enquiry is made. A specific instruction can be stored in the instructions, e.g.:
   
   ‘Only perform the live chat follow-up action if the cost of product X is requested.’ This rule ensures that the follow-up is only triggered in this specific case. How to create instructions is described in this article.
2. With placeholders in the query
   Scenario: An AI agent for cost questions should forward to an order form. To ensure that the yes-no-question refers to the requested product, a placeholder is used in the action description. The wording of the query is rule-based and adapted to the respective product type.
   
   Result for the question ‘How much does the business package cost?’
   

AI agent settings

AI agent settings relate in particular to the persona and certain communicative behaviours or characteristics of the AI agent. The specific functionality and corresponding setting options are described in this article.