AI Chat labeling

The AI Chat Labeling feature allows you to automatically extract crucial metrics and insights from your chat conversations using Artificial Intelligence.

Overview

By configuring metrics, you can send chat transcripts as prompts to a dedicated AI model, which then analyzes the conversation and determines the correct value (e.g., a classification, a label, a number, or a piece of text) based on your specific instructions.

This powerful tool enables automated data capture, enriching your reports and saving countless hours of manual review.

Key Concepts and Components

Metrics

A metric defines a specific piece of information to be extracted or calculated from a chat. Each metric is configured with a type, a data source, and specific instructions for the AI.

AI Labeling Tokens

Processing a chat transcript with the AI consumes AI Labeling Tokens.

• All customers receive a default allocation of tokens.
• For customers with high chat volume, the default allocation may not cover all chats.
• Additional tokens can be purchased to ensure comprehensive labeling across all conversations.
• To manage token usage, metrics can be configured to exclude chats in specific rooms (see Configuration Options).

Data Sources

Metrics can pull data from two sources:

Metric Types

The system supports five distinct metric types, determining the format of the output the AI (or API) provides:

Metric Type	Description	AI Output Constraint	Example Use Case
Classification	The AI must select exactly one value from a predefined list of classes.	Single Choice	Determine the intent of the chat (e.g., Sales, Support, General Inquiry).
Labeling	The AI can select zero, one, or multiple relevant values from a predefined list of labels.	Multi-Choice	Identify all products discussed (e.g., 'Product A', 'Product B', 'Service C').
Tagging	The AI can freely choose zero, one, or multiple tags that it deems relevant to the conversation.	Freeform Multi-Choice	Capture sentiment or key themes (e.g., 'Frustrated', 'Positive Feedback', 'Pricing Question').
Number	The AI provides a single numeric value.	Numeric Value	Extract the estimated deal size in a sales chat or a rating score.
Text	The AI provides a textual value.	Freeform Text	Summarize the key action taken or a required follow-up.

Creating and Configuring an AI Chat Labeling Metric

AI Chat Labeling configuration can be accessed by going to Giosg reporting dashboard at https://reporting.giosg.com and clicking “Reporting settings” from under the three-dot menu on the top right corner of the page.

To create a new AI Chat Labeling metric, you will need to provide the following essential information:

 

1. General Settings

Field	Requirement	Description
Name	Required	A human-readable name for the metric (e.g., "Chat Intent Classification", "Customer Sentiment").
Metric Type	Required	Select one of the five types: Classification, Labeling, Tagging, Number, or Text.
Data Source	Required	Choose AI (Default) or API.

 

2. Description for AI (The Prompt)

This section is crucial as it instructs the AI on exactly what to do with the chat log.

Field	Requirement	Description
Name (for AI)	Required	A short, descriptive name the AI uses internally. It should help the AI understand its core task (e.g., "CLASSIFY_INTENT", "FIND_PRODUCTS").
Description (The Prompt)	Required	This is the prompt sent to the AI. Clearly explain what is expected, any special rules, contextual information, and the required output format.

 

Example Prompt Tip: "Your task is to classify the customer's primary intent. Analyze the full chat log. If the intent is unclear, classify it as 'General Inquiry'. Provide only the single most relevant class name from the provided list."

 

3. Type-Specific Configuration

Depending on your selected Metric Type, additional fields will be required to constrain the AI's output:

Metric Type	Required Field	Description
Classification	Classification Classes	A list of predefined values. The AI must select exactly one from this list.
Labeling	List of Labels	A list of predefined values. The AI can select zero, one, or multiple values from this list.

 

4. Excluded rooms

To help manage your usage of AI Labeling Tokens, you can exclude specific conversations from being processed:

• Exclude Chats in Rooms: Select one or more chat rooms where the metric should not be applied. Chats originating from these rooms will be skipped, thus saving tokens. This is recommended for rooms that generate low-value or high-volume, repetitive chats and testing rooms.

Viewing Results in the Reporting Dashboard

Once your AI Chat Labeling metric is created and actively processing data (or receiving data via API), you can add it to your own Reporting Dashboard for visualization and analysis.

1. Navigate to your Reporting Dashboard at https://reporting.giosg.com/my-dashboard
2. Add Cards: Locate the three-dot menu (...) on the top right corner of the dashboard page.
3. Select "Add cards" from the menu.
4. Choose Metrics: A list of available metric cards will appear. Select the AI Chat Labeling metric(s) you just created and click the plus icon.
5. Visualization: You can choose the visualization type for your card that you think is the most suitable. Note that not all visualization types are available for all metric types.
6. The card will be added to your dashboard, allowing you to view the results, trends, and data distribution based on the labels and values collected.
7. Note that the dashboards are user specific so you can only add the metrics you are interested in.

 

Open the list of metrics:

 

Select the metric(s) to add:

Select the visualization type for the results:

 

Data Source: API Integration

When the Data Source is set to API, the metric's value is provided by an external system rather than being calculated by the AI. This is ideal for integrating existing business data (like an NPS score or lead quality score) into your reporting alongside your chat data.

1. Configuration and API Key

When you set the Data Source to API for a metric, the system generates a unique endpoint for you to post data to.

The bottom of the metric editing page will display the required API endpoint and an example of the expected JSON payload.

Note: The unique Metric ID and the API Endpoint URL will be displayed in this section. The endpoint is secured and requires an API Key authentication.

2. Submitting Metric Data

To update a metric's value for a specific chat, you must make a POST request to the provided API Endpoint. Provided JSON payload can be copied from the bottom of the page. Note that the exact payload structure depends on the type of the metric.