How to Configure Custom AI Skills for AI Employees?

I. Function Introduction

Custom AI skills can be configured into a set of "trigger conditions + execution instructions" for specific business scenarios. When a customer inquiry meets the conditions, the AI employee will handle the current request according to the corresponding skill.

Skills can combine variables, knowledge, and tools to complete one or more processing steps. For scenarios with clear triggering conditions and processing procedures, such as after-sales applications, lead screening, and document delivery, separate skills can be created for easy review and adjustment later.

This article's link: "How to Configure Custom AI Skills for AI Employees?"

II. Creating Custom AI Skills

Go to the "Configuration" page of the corresponding AI employee, find "Custom AI Skills" in the "Skills" area, click "Settings", and then click "Create Skill".

1. Fill in the skill name

Please enter a name that clearly identifies the current purpose of the skill. It is recommended to use a name that reflects the specific business scenario, which will facilitate subsequent management and adjustments.

2. Set the "Trigger Conditions"

"Trigger conditions" are used to tell the AI employee when to use the current skill .

It is recommended to focus on describing the customer's needs, intent, or current session status, rather than filling in specific processing steps here.

If skill activation requires further evaluation based on customer or conversation information, you can click "Insert Variable" to use contact attributes, custom attributes, or system attributes. For detailed instructions, please refer to "Using Variables and Tools in Custom AI Skills" below.

3. Enter the "Execution Instruction"

"Execution Instructions" are used to explain how the AI employee should handle the current needs after the skill is triggered, including responding to the request, referring to the relevant knowledge, and performing the necessary actions.

If a skill requires multiple steps, multiple operations can be set up according to the actual business process.

For example, the "After-sales Request Processing" skill in the image below will first check the after-sales rules, then collect the order number required to process the after-sales request, and finally transfer the session to a human for further processing.

After completing the configuration, click "Save".

III. Using Variables and Tools in Custom AI Skills

1. Inserting variables

By clicking "Insert Variable" in "Trigger Conditions" or "Insert Variable/Tool" in "Execute Instructions", you can reference information from the current client or session, allowing AI to make judgments and handle situations based on the actual circumstances.

Currently, contact attributes, custom attributes, and system attributes are available, with system attributes supporting "AI reply rounds".

For example, you can refer to the following configuration method:

2. Utilize tools and knowledge

In the "Execute Instruction" menu, clicking "Insert Variable/Tool" allows you to insert tools or specify the knowledge content to be used.

Different functions can be used in the following ways:

• Before using "Switch to Human Assistant", you need to enable and complete the configuration for "Switch to Human Assistant", otherwise the skill will not be able to be used normally.
• When invoking "Attribute Collection," you need to select the specific attribute or tag that has already been created. This tool is used to record information already provided by the customer; if the customer has not yet provided the corresponding information, you need to first instruct the AI to ask the customer in the execution command.
• When you call "Send Knowledge Base Material", you need to select the specific content that has been uploaded to the Knowledge Base "Material".

Related tutorials: "How to Set Up AI Employees to Switch to Human Agents?" , "How to Set Up AI Employees to Collect Customer Information?" , "How to Set Up AI Employees to Update Customer Tags?" , "How to Use Materials to Get AI Employees to Respond with Images, Audio/Video, and Attachments?"

IV. Testing and Release

After completing the skill configuration, it is recommended to first simulate a real customer consultation through the "Q&A Test" on the right side of the page to check whether the skills are triggered and executed as expected.

During testing, you can input questions about whether the current skill should be triggered or not, and focus on checking the following:

• Whether the skill is triggered in the correct scenario;
• Whether the inserted variable is used correctly;
• Whether the operations such as attribute collection, knowledge retrieval, material delivery, and transfer to human assistant meet expectations.

(Note: In the "Q&A Test," you can check if the AI calls the "Transfer to Human Agent" tool as expected, but the test will not actually execute the transfer to a human agent. If you need to verify the effectiveness of transferring to a human agent in a real conversation, you can simulate actual reception through chat plugins or other methods. Related tutorial: "How to Test the Actual Reception Effectiveness of AI Employees?" .)

If the results are not as expected, you can adjust the "triggering conditions" or "execution instructions" and then test again.

After confirming that everything is correct, click "Publish" in the upper right corner of the page to apply the latest configuration to the actual reception.

V. Frequently Asked Questions

1. Why are the configured skills not triggering as expected?

First, check whether the "triggering conditions" accurately describe the actual customer needs.

If the skill is not triggered, is triggered accidentally, or the execution result is not as expected, you can click "Not good answer? Troubleshoot the reason" below the answer in "Q&A Test" to use the troubleshooting assistant to view possible reasons and optimization suggestions.

After adjusting the "triggering conditions" or "execution commands" based on the investigation results, re-verify them through "question and answer testing." Only release and use it after confirming that the effect meets expectations.

Related tutorial: "How to troubleshoot when AI employee responses do not meet expectations?"