Smart Argument Logic

​

Specifications

Basic Settings

​

1. Define Specifications

Here the foundation for the AI agent is laid. This includes:

• Name: The name of the agent should reflect the area of responsibility, e.g. “Instalment Agent”.

• Description: A brief description helps to grasp the purpose of the agent at a glance.

• Topic: The topic arises from the existing skill tags. For example, an instalment agent could be assigned to the skill tag “Payments”. The assignment ensures clear organizational assignment and, for example, facilitates access via the template catalog.

• Channels: Determine through which communication channels the agent should act - Email, chat, letter, phone, etc. An agent can easily be assigned to multiple channels.

​

2. Choose Avatar

The avatar gives the AI agent a visual identity and emphasizes its individual character. Various options are available:

• Style: Should the avatar appear more professional, friendly, or neutral?

• Color Accents: In addition, colors can be selected that, for example, match the brand identity.

• Recognition Value: A uniform avatar across all channels strengthens the visual appearance of the agent.

​

3. Configure Intelligence

This decides how the agent solves tasks:

• Smart Argument Logic: Rules in natural language without programming - ideal for standard processes.

• Rule-based Logic: Code-based workflows for more complex procedures.

​

4. Customize Personality

The agent’s personality defines how it communicates - not only in terms of content, but also in tone. Communication style can be finely adjusted via intuitive sliders:

• Informal to Formal: Should the agent appear relaxed or businesslike?

• Speaking Style: From brief answers to detailed explanations.

• Concise to extensive: Determines the depth of detail of the feedback.

These adjustments ensure the agent fits perfectly with the company’s language and positively supports the customer experience.

​

Detection

Methods and Configuration

​

Authentication

If this option is activated, requests from non-authenticated customers will be ignored by the AI agent. This setting is particularly suitable for processes involving sensitive data or requiring customer verification.

​

Detection Methods

Various methods are available for assigning AI agents:

​

Instructions

Prompt, Tools and Techniques

​

Example Prompt of a Tariff Advisor Agent

You are an expert in reclaiming customers who have terminated their contracts for utilities providers. As a tariff advisor, you want to effectively, yet politely inspire the customers over the phone to opt for an affordable tariff instead of going through with the cancellation. You have called the customer. Please proceed as follows:

1. If the customer says the termination was incorrect, note that in cancel_termination and withdraw the termination.

2. Otherwise, ask why the customer has terminated their contract. Record the reason in document_reason_for_termination.

3. Next, ask if you can make a prepared change offer to save them money. Address the reason for cancellation, if relevant. Accept it and end the call if they decline.

4. If the customer is interested, retrieve the offer with get_tariff_offer. Briefly present the offer by focusing on the highlights. Phrase numbers in words (e.g., 1.5 → one point five). Use natural language, not JSON variables or abbreviations like kWh. If the customer is not interested, end the call.

5. Answer any questions from the customer about this tariff on request. If the customer wants the tariff sent by email, use send_offer_by_email and confirm the dispatch, mentioning the stated email address.

6. Ask if the customer would like to accept the offer. If they agree, initiate the tariff change with do_tariff_change. Otherwise, accept the decision and end the call.

7. Confirm a successful switch to the customer and end the call.

Focus exclusively on this process and do not address other topics. You already know the customer, identification is not necessary. Use the formal “you” in addressing the customer.

​

Custom Tools

These tools assist the AI agent in carrying out processes and instructions efficiently and effectively. Depending on the configuration of these tools, the actions of the AI agent are triggered, adjusted on the basis of identified customer needs, and completed in the defined course of the process.

In the example above, these tools are being used:

• Cancel Termination: Revokes an existing termination.

• Document Reason for Termination: Records the customer’s reason for termination.

• Get Tariff Offer: Fetches an individually tailored tariff offer.

• Do Tariff Change: Executes the tariff change once the customer has decided on an offer.

• Send Offer by Email: Sends the created tariff offer directly to the customer’s email address.

More information on Custom Tools and how to configure them can be found in the User-Defined Functions / Tools section.

​

Integrated Tools

These tools extend the capabilities of the AI agent by facilitating access to relevant data sources and supporting standardized processes. They are pre-configured and ready for immediate use.

• Retrieve Knowledge Database entries: Provides access to the contents of wiki articles.

• Retrieve Customer and Contract data: Grants access to customer and contract data.

• Retrieve Response Examples: Offers access to responses to similar customer queries from the past.

• Retrieve Customer Ticket History: Displays the customer’s ticket history.

​

Testing and Publishing

Operating Mechanism

​

Test Procedure

• A real ticket ID is selected and added.

• The AI agent processes the ticket based on the defined logic.

• The result is displayed as successful or failed.

​

Result Interpretation

• Successful: The test confirms that the AI agent is responding as intended. The AI agent can now be published.

• Failed: In this case, the existing settings need to be examined. Instructions, detections, input parameters, business logic, and/or output handling should be investigated for possible inconsistencies. After making the adjustment, the test should be run again.