Dashboard data export

Overview

This documentation explains how data is exported from the InsurDirect dashboard, based on data source configurations and attribute mappings.

How Export works

The export always reflects the data currently visible on the dashboard, including any filters applied (for example, filtering on a specific week).

Formatting Specific Data Types

Conversation type data

Conversation-type fields (such as "chat_conversation", "email_conversation", "audio_call") are transformed into structured character strings that contain the entire exchange, using this syntax:


agent: Hello and welcome to Un souriANT.
customer: Can you help me? I'd like to know the agency's opening hours.
agent: The Lille agency is open from 9am to 5pm.
customer: Thank you very much!


Type Date data

An evolution is currently underway concerning date fields:

• In the near future, all dates will be converted to the local time of the person who performed the export.
• The format will include the complete timezone (for example: 2025-04-17 10:30:00+02:00 for an export made in Paris in summer time)
• This enhancement will make it easier to analyze temporal data in different geographical contexts

Case 1: Export with a single data source

When only one data source is configured in the dashboard:

• Export includes all original data source fields
• AI-analyzed fields ("text", "chat_conversation", "email_conversation", "audio_call") generate additional columns
• Each analysis column bears the name of the original field followed by ".topics" (example: "conversation.topics").
• These columns contain the business topics identified and the associated sentiment

Case 2: Export with multiple data sources

When several data sources are mapped in the dashboard:

• Export includes fields defined in attribute mapping
• The field names correspond to those chosen manually in the configuration.
• AI analyses are also included, using the same naming format (.topics).

Attribute mapping configuration

Attribute mapping determines the export structure when multiple sources are used:

1. In the "General parameters" section, define the included data sources
2. In "Attribute mapping", associate the fields of each source with a common column name.
3. Common attributes (such as "Customer", "Conversation", "Date") are mapped to source-specific fields

‍

Exported data structure

The final export takes the form of a table with:

• A unique ID column for each entry
• The original data source (data_source)
• Creation timestamp (created_at)
• Values of mapped attributes
• IA analysis columns with .topics extension

Understanding topic_levels in AI Analysis

The "topic_levels" mentioned in the IA analysis columns correspond to the different levels of the classification plan:

• level_1Represents the main subject (e.g. "Product", "People")
• level_2Corresponds to a more specific sub-topic (e.g. "Reimbursement", "Schedules").
• level_3Can appear for even more precise subjects if configured

Each level is associated with a sentiment analysis ("positive", "negative", "neutral") that reflects the tone of exchanges on that specific subject.

Example of structured analysis:


json
{
 "topic_levels": [
   {"level_1": "Produit", "sentiment": "positif"},
   {"level_2": "Remboursement", "sentiment": "neutre"},
   {"level_1": "Personnel", "sentiment": "positif"}
 ]
}



Concrete examples

Example with the InsurDirect source - Callbot

For Callbot data, export includes:

• Identification information (ID, timestamp)
• The content of the conversation
• Analysis of topics covered (topic_levels)
• Customer contact information
• Reason for contact and unanswered questions

Example with multiple sources (Post call, WebApp, Callbot)

In this case, the export unifies the data with:

• A common structure based on mapping
• Source-specific attributes converted to standardized format
• Harmonized subject analyses in a single format

Good Practices

• Check your attribute mapping before export to ensure data consistency
• Use explicit attribute names to facilitate subsequent analysis
• Make sure all important text fields are correctly mapped to benefit from AI analysis.

‍