In this quickstart, you'll learn how to build a comprehensive customer service analytics system that processes audio, text, and document data using Snowflake Cortex AI functions. This application demonstrates how to extract insights from multimodal data sources including call recordings, chat logs, support tickets, and PDF documents.

A production-ready customer service analytics system that:

1. Download the setup.sql file from GitHub
2. Open a new SQL worksheet in Snowflake
3. Copy and paste the contents of setup.sql into the worksheet
4. Run the entire script to create:
   • Database and schema
   • Storage stages for audio files and documents
   • Required tables for storing results
   • Sample data tables

1. Download the Notebook from GitHub (NOTE: Do NOT right-click to download.)
2. In your Snowflake account:
   • On the left hand navigation menu, click on Projects » Notebooks
   • On the top right, click on Notebook down arrow and select Import .ipynb file from the dropdown menu
   • Select the file you downloaded in step 1 above
3. In the Create Notebook popup:
   • For Notebook location, select MULTIMODAL_CUSTOMER_SERVICE for your database and DATA as your schema
   • Select your Warehouse
   • Click on Create button

The Notebook demonstrates how to chain multiple Snowflake Cortex AI functions together to analyze customer service calls. This section covers the complete audio processing pipeline.

The first cell shows a comprehensive query that combines multiple AI functions:

1. AI_TRANSCRIBE - Converts audio files to text with speaker identification
2. Flatten and Combine - Extracts transcript segments and creates full conversation text
3. AI_TRANSLATE - Automatically translates conversations to English
4. AI_SENTIMENT - Analyzes emotional tone (positive, negative, neutral, mixed)
5. AI_CLASSIFY - Categorizes calls into custom issue types
6. AI_COMPLETE - Generates concise 50-word summaries using claude-sonnet-4-5

This unified approach processes each call through the entire analytics pipeline in one execution, storing comprehensive results in the transcription_results table.

The Notebook then breaks down each AI function step-by-step, creating temporary tables to demonstrate:

The AI_CLASSIFY function uses five business-specific categories:

1. Fraud & Security Issues - Unauthorized transactions, identity theft, account freezes
2. Technical & System Errors - Login problems, system glitches, auto-pay failures
3. Payment & Transaction Problems - Duplicate charges, failed payments, fee disputes
4. Account Changes & Modifications - Policy changes, fund transfers, coverage adjustments
5. General Inquiries & Information Requests - Status checks, documentation requests

Each category includes detailed descriptions to ensure accurate AI classification.

After processing, query the transcription_results table to see all analyzed calls with transcriptions, translations, sentiments, categories, and summaries. The Notebook includes cleanup steps to drop temporary tables.

The Notebook demonstrates AI_PARSE_DOCUMENT for extracting structured information from PDF files. The function processes all documents in the COMPANY_DOCUMENTS stage using 'LAYOUT' mode with page splitting enabled.

The function returns detailed JSON containing:

This parsed data can be used for document search, information extraction, or integration with RAG systems.

For chat logs, the Notebook implements a validation system using AI_EXTRACT, AI_CLASSIFY, and AI_SENTIMENT to check data quality:

1. Reconstruct Conversations - Flatten message arrays into complete text
2. Apply AI Analysis - Analyze conversations with multiple AI functions
3. Validate Self-Reported Data - Compare AI results against agent classifications
4. Flag Discrepancies - Identify mismatches in categories and sentiments

The AI_EXTRACT function extracts structured information like:

The final analysis uses AI_COMPLETE to semantically compare support tickets with their corresponding chat logs. The AI analyzes both data sources and returns:

The query creates a comprehensive flagging system that identifies issue misalignments, category mismatches, and product inconsistencies. This cross-validation ensures data consistency across customer service systems and helps identify where processes need improvement.

Congratulations! You've successfully built a comprehensive multimodal customer service analytics system using Snowflake Cortex AI functions. Using Snowflake Notebooks, you've implemented a solution that transcribes audio calls, translates conversations, analyzes sentiment, categorizes issues, generates summaries, parses documents, and validates data quality - all while keeping your data secure within Snowflake's environment.

Documentation:

Sample Code & Guides: