# Understanding Conversation Intelligence > Source: https://help.kustomer.com/en_us/understanding-conversation-intelligence-r1aBEjuYzl Last updated: 2026-10-16T18:23:39.992Z When customers interact with your team across any channel, their conversations accumulate a wealth of information: why they reached out, what happened during the interaction, how satisfied they seemed, and other details that can help you understand and serve them better. AI attributes turn that unstructured information into structured data, powering Conversation Intelligence insights, reporting, and actions that you can use throughout Kustomer. ### Understanding how AI attributes power Conversation Intelligence [Conversation Intelligence](https://help.kustomer.com/en_us/understanding-conversation-intelligence-r1aBEjuYzl) uses AI to evaluate information in Kustomer conversations across any channel, and automatically populate fields with the results. Once AI attributes have stored this structured data, Conversation Intelligence uses it in workflows, routing, reporting, and other parts of Kustomer. There are three types of Conversation Intelligence: * **Classifiers** categorize information into predefined options. * **Scores** evaluate information against criteria and return a numeric score. * **Extractors** identify and save specific information as structured data. Which type you choose depends on what you want to learn from your customer interactions. ### Categorize conversations with classifiers Use a **classifier** when you want AI to place a conversation into one or more predefined categories. You define the categories the AI can choose from, along with descriptions that explain what each category means. The AI evaluates the available context and selects the category that best matches it. For example, you could create a **Contact reason** classifier with categories such as: * Order status * Return or exchange * Cancel an order * Product question * Payment issue If a customer writes, “Where is my order?”, the classifier can categorize the conversation as **Order status** giving you a consistent contact reason without requiring an agent to tag the conversation manually. You can use classifier outputs to understand why customers contact you, route conversations, trigger workflows, or segment conversations for reporting and analysis. Classifiers can support: * Single-select, where only one label is chosen from a list of possible options. * Multi-select, where multiple labels can be chosen based on which labels fit the description and content of the conversation. * Hierarchical categories, where labels can be nested beneath other labels depending on how you want to organize your taxonomy. **Note:** Multi-select is not possible with nested categories. ### Score interactions on a numeric scale Use a **score** when you want AI to evaluate an interaction against a set of criteria. You define the criteria that matter, describe what each criterion means, and assign each one a weight. AI evaluates the available context, and produces a single weighted score (0-100 range). For example, you could create an AI-inferred satisfaction score using criteria such as: * How useful was the automation's first response? * How well did the automation understand the issue? * How successful was the interaction with the customer? * How well did the automation serve the customer according to your company values? You could also create scores for things such as policy compliance or adherence to a specific support process. In addition to the numeric result, score attributes include a rationale explaining how the evaluation was reached. This gives you both a value you can use programmatically and context for understanding that value. ### Extract specific information Use an **extractor** when you want AI to find a specific piece of information and save it as structured data. Instead of giving AI a predefined set of categories, you provide instructions describing what information it should identify. For example, an extractor could identify: * A product mentioned by the customer * An order number * A competitor name * A requested cancellation reason * A promised follow-up date * A concise summary of the customer's issue You define what the extractor should look for and the type of data it should return. AI evaluates the available context, extracts the relevant information, and stores the result in the attribute. Extractor outputs can be text, longer text, numbers, booleans, dates, URLs, email addresses, phone numbers, or lists of text values, depending on the information you want to capture. ### Using Conversation Intelligence in Kustomer The value of Conversation Intelligence isn't limited to displaying information on a conversation. Because the result is stored as structured data, it can be used across all conversation channels and in other parts of Kustomer. For example, you might: * Route conversations classified as **Cancellation request** to a retention team. * Trigger a workflow when an interaction quality score falls below a certain threshold. * Report on the most common reasons customers contact your team. * Extract product names and use them to analyze which products generate the most support volume. * Identify information that agents would otherwise need to read through an entire conversation to find. In this way, Conversation Intelligence, powered by AI attributes, creates a bridge between the unstructured information contained in customer interactions and the structured data your team can use to automate work, make decisions, and understand customer trends.