1. Introduction/Overview: AI and the Quest for Meaning
This guide explains effective "prompt" (input/definition) creation methods to ensure the system best understands your inputs and maximizes the accuracy of your analyses when creating user-defined categories and subcategories for AI-powered call analysis in the Call Center Studio CX Insights feature. Successful prompts achieve validation scores of 80% or higher, ensuring your categories are used correctly in analyses.
Why Does AI Need Detailed Inputs?
Advanced AI models, especially Large Language Models (LLMs) used in systems like Call Center Studio CX Insights, don't "understand" human language in the way humans do; instead, they operate by recognizing patterns and relationships learned from vast datasets. Unlike the human brain's intuitive and contextual understanding, meaning for AI largely depends on the clarity and specificity of the data you provide.
This is where your "prompts" come into play:
Focusing and Guiding: AI has a vast universe of knowledge learned during its general training. However, your operation's specific workflows, terminology, and priorities may differ from this general knowledge. Detailed prompts help focus the AI's attention on the specific topics, phrases, and intents important to you. Like a spotlight, they illuminate the area that needs to be analyzed.
Defining Boundaries: Call center conversations are often complex and multi-layered. A single conversation can touch on multiple topics. Clearly defining the boundaries of your category definitions with "Correct Examples" and especially "Incorrect Examples" helps the AI distinguish between phrases that may seem similar but carry different meanings. This prevents "almost correct" but actually wrong interpretations.
Teaching Specific Context: Your business might have unique product names, campaign codes, process steps, or customer segments. The AI can only accurately incorporate this specific context into its analyses if you teach it. Your prompts facilitate the transfer of this "corporate language" and business logic to the AI.
Modeling Human Expertise: The ultimate goal is for the AI to categorize your calls like an experienced human analyst would. The prompts you input serve as "training material" that transfers this human expertise and knowledge to the AI. The AI processes these inputs to learn your defined perception of "correct" and "incorrect" and uses this learning in subsequent analyses.
In short, AI "sees" and interprets the world according to the rules and examples you provide. The clearer, more consistent, and comprehensive your prompts are, the more accurate, meaningful, and valuable the AI's analyses will be for your business. This isn't about the AI "thinking," but about it matching and classifying complex statistical patterns within the framework you provide.
2. Category and Subcategory Definition Fields and Their Importance
Each field you fill when defining your categories and subcategories helps the AI understand the purpose, scope, and boundaries of that category. The clarity and accuracy of the information you enter directly impact the quality of your analysis results.
2.1. Main Category Definition Fields
When manually creating a new main category, you should pay attention to the following fields:
Category Name:
Purpose: This is the short and clear title of the category that will appear in the system and reports.
Tips:
Choose a name that is understandable and memorable.
It should reflect the content of the category.
There is a maximum limit of 30 characters.
Example: "Product Appreciation," "Service Satisfaction," "Positive Feedback"
Prompt Input (Category Description):
Purpose: This explains in detail to the AI what this category is about and what types of conversations or topics it should include.
Tips:
Clearly define the purpose and scope of the category.
Specify which situations should be included in this category and which should not.
Be as specific as possible.
There is a maximum limit of 1000 characters.
Example (for Product Appreciation): "Covers situations where the customer expresses positive thoughts about a product they use, or satisfaction with its features, quality, or performance. Specifically includes praise and appreciation directed at the product itself. General service satisfaction or positive comments about personnel do not fall into this category."
Correct Examples:
Purpose: Provides the AI with real conversation sentences or phrases that exactly match the category defined in the "Prompt Input." It materializes what kind of texts the model should assign to this category.
Tips:
Choose realistic examples that well represent the category.
You can add a maximum of 2 examples.
Each example can be a maximum of 125 characters.
At least one example must be entered in the "Correct Examples" and "Incorrect Examples" fields.
Example (for Product Appreciation):
"The camera on the new phone I bought is truly amazing, I love it."
"This software has made our work so much easier, it's a fantastic product."
Incorrect Examples:
Purpose: Provides the AI with conversation sentences or phrases that might resemble the category definition but do not actually belong to it. It teaches the model what kind of texts it should not assign to this category, thus clarifying the boundaries.
Tips:
Choose examples that could be confused with the category, "almost correct" but actually wrong.
You can add a maximum of 2 examples.
Each example can be a maximum of 125 characters.
At least one example must be entered in the "Correct Examples" and "Incorrect Examples" fields.
Example (for Product Appreciation):
"Your customer representative was very helpful." (This is personnel satisfaction, not product appreciation)
"The delivery arrived faster than I expected, thanks." (This is service satisfaction, not specific product appreciation)
2.2. Subcategory Definition Fields
When creating a subcategory linked to a main category, you should pay attention to the following fields:
Subcategory Name:
Purpose: This is the short and clear title of the subcategory that will appear under the main category.
Tips:
Choose a name that reflects its relationship with the main category and its own specific focus.
There is a maximum limit of 30 characters.
Example (Main Category: "Service Satisfaction," Subcategory: "Thanks for Quick Resolution")
Prompt Input (Subcategory Description):
Purpose: This explains in detail to the AI what this subcategory is about and which specific topic within the main category it addresses.
Tips:
Clearly define how the subcategory differs from the main category and its own specific scope.
There is a maximum limit of 500 characters.
Example (Subcategory: "Thanks for Quick Resolution," Main Category: "Service Satisfaction"): "Covers situations where the customer expresses satisfaction and thanks for an issue or request being resolved faster than expected. Emphasis is specifically on the speed of the resolution. General appreciation for service quality does not fall into this subcategory."
Correct Examples:
Purpose: Provides the AI with examples that perfectly match the subcategory definition.
Tips:
Choose specific examples that reflect the particular focus of the subcategory.
You can add a maximum of 2 examples.
Each example can be a maximum of 125 characters.
At least one example must be entered in the "Correct Examples" and "Incorrect Examples" fields.
Example (for Subcategory: "Thanks for Quick Resolution"):
"Thank you so much for resolving my issue this quickly."
"Wow, amazing job, my request was addressed and handled instantly."
Incorrect Examples:
Purpose: Provides the AI with examples that might be confused with the subcategory but do not actually belong to it.
Tips:
Choose examples that might belong to other subcategories of the main category or to the general main category.
You can add a maximum of 2 examples.
Each example can be a maximum of 125 characters.
At least one example must be entered in the "Correct Examples" and "Incorrect Examples" fields.
Example (for Subcategory: "Thanks for Quick Resolution"):
"Your customer representative was very polite." (This is satisfaction with personnel behavior, not resolution speed)
"I am very pleased with the overall service quality of your company." (This is a general statement, not specifically focused on quick resolution)
3. Validation Process and Score Improvement
After your input definitions and examples are saved, they are subjected to a validation process by the AI and receive a score between 1 and 100.
For your category or subcategory to be used in analyses and published, it is aimed to reach a validation score of 80% or higher.
In Case of a Low Validation Score:
If your score is below 80%, you will receive a warning in the interface like: "Validation score low. Calls will not be analyzed for this category/subcategory... increase the validation score to 80. You can use the suggestions below to increase the validation score."
The system will offer AI-generated contextual suggestions to help you increase your score. Carefully review these suggestions and make revisions in the "Prompt Input," "Correct Examples," and "Incorrect Examples" fields.
Making your definitions clearer and choosing more specific and distinctive examples can help increase your score.
4. General Tips and Best Practices
Be Clear and Understandable: Remember that AI does not operate like a human but according to the definitions and examples you provide. Use expressions that are as clear, unambiguous, and straightforward as possible.
Be Specific: Instead of general statements, make specific definitions indicating exactly what the category or subcategory includes and excludes.
Use Keywords: Make an effort to use important keywords related to the category or subcategory in the "Prompt Input" and examples.
Be Consistent: The definition and examples you provide for a category or subcategory should be consistent within themselves.
Define Boundaries: Clearly draw the boundaries of the category, especially by using "Incorrect Examples." This helps the AI correctly distinguish between similar but different topics.
Iterative Approach: You may not achieve the perfect result on your first try. Create your inputs, check the validation score, consider the system's suggestions, and if necessary, try to increase your score by updating your definitions and examples. This is a process of trial-and-error and refinement.
Pay Attention to Character Limits: Be careful not to exceed the specified character limits for each field.
By following the recommendations in this guide, you can create more accurate and effective user-defined categories in Call Center Studio CX Insights and get the most value from your call analyses. For setup details, you can review our CX Insights User-Defined Categories guide.