Processing Call Center Transcripts with Tonic Textual

The Setup

You're Eileen, the Analytics Manager at Initech. Your boss Bill Lumbergh just walked into your office:

"Yeah... I'm gonna need you to analyze all our call center recordings for compliance. But legal says we can't use real customer data. That'd be great."

The challenge: Thousands of call transcripts with customer names, SSNs, credit cards. You need to remove PII but keep conversations understandable for quality analysis.

What We'll Build

A transcript processing system that:

• Redacts PII from call transcripts automatically
• Preserves conversation context and meaning
• Generates safe transcripts for QA and training
• Works with any transcript format (JSON, CSV, plain text)

Prerequisites

# Install Tonic Textual SDK
pip install tonic-textual python-dotenv pandas

# Create environment file
echo "TONIC_TEXTUAL_API_KEY=your_api_key_here" > .env

Step 1: Process Your First Call Transcript

import os
import json
from dotenv import load_dotenv
from tonic_textual.redact_api import TextualNer

# Load API key
load_dotenv()
tonic = TextualNer(api_key=os.getenv("TONIC_TEXTUAL_API_KEY"))

# Sample transcript from a call recording
call_transcript = {
    "call_id": "CALL-2024-001",
    "timestamp": "2024-03-15T14:30:00Z",
    "agent": "Bruce Sun",
    "duration": 245,
    "transcript": """
Agent: Thank you for calling Initech, my name is Bruce. How may I assist you today?

Customer: Hi Bruce, I'm Milton Waddams. I need to update my payment method.
My old card number was 4532-1234-5678-9012 and the new one is 4916-3385-0987-6543.
My account number is ACC-78234.

Agent: Thank you Mr. Waddams. I can help you with that.
Can you verify your SSN for security?

Customer: Sure, it's 987-65-4321.

Agent: Perfect. I've updated your payment method.
Is there anything else I can help you with?

Customer: No, that's all. Thank you!

Agent: Have a great day!
    """
}

print(f"Processing transcript for call {call_transcript['call_id']}")
print(f"Duration: {call_transcript['duration']} seconds")
print(f"Original preview: {call_transcript['transcript'][:150]}...")

Step 2: Remove All PII for Compliance

Now let's automatically redact all PII from the transcript:

# Convert to JSON for processing
transcript_json = json.dumps(call_transcript, indent=2)

# Redact sensitive information
redacted_response = tonic.redact(transcript_json)

# Parse the redacted result
redacted_data = json.loads(redacted_response.redacted_text)

print("\n" + "="*50)
print("BEFORE - Original Transcript:")
print(call_transcript['transcript'][:300])

print("\n" + "="*50)
print("AFTER - Redacted Transcript:")
print(redacted_data['transcript'][:300])

# The redacted version will have:
# - Names replaced with consistent tokens
# - Credit card numbers removed
# - SSN removed
# - Email addresses redacted
# - Account numbers redacted
# - But conversation flow preserved!

Step 3: Process Multiple Transcripts at Scale

Bill Lumbergh walks by again: "Oh, and if you could process ALL the call transcripts from this week... That'd be great."

No problem - let's batch process:

# Multiple transcripts to process
transcripts = [
    {
        "call_id": "CALL-001",
        "agent": "Bruce Sun",
        "transcript": "Customer Robert Chen called about account ACC-78234. Card 4532-1234-5678-9012."
    },
    {
        "call_id": "CALL-002",
        "agent": "Mike Johnson",
        "transcript": "Lisa Wang needs help. Email: lwang@email.com. Phone: 555-0123."
    },
    {
        "call_id": "CALL-003",
        "agent": "Bruce Sun",  # Same agent as CALL-001
        "transcript": "Robert Chen calling back about account ACC-78234."  # Same customer
    }
]

# Process each transcript
redacted_transcripts = []
for transcript in transcripts:
    # Redact PII
    response = tonic.redact(json.dumps(transcript))
    redacted = json.loads(response.redacted_text)
    redacted_transcripts.append(redacted)
    print(f"Processed {transcript['call_id']}")

# Notice: Same names get same tokens for consistency!
print("\nConsistency check:")
print(f"Call 1 agent: {redacted_transcripts[0]['agent']}")
print(f"Call 3 agent: {redacted_transcripts[2]['agent']}")
print("Same person = same token across all calls!")

Victory: Ship It to Legal!

You walk into Bill's office with confidence:

"Bill, we've processed all call transcripts. Zero PII exposure. Legal approved it. Quality team can start their analysis."

Bill actually smiles "Great. Oh, and I'm gonna need you to come in on Saturday..."

"Just kidding. Nice work!"

Results

Here's what you've accomplished:

• ✅ Privacy Protection: 100% PII removal - Legal loves you
• ✅ Data Utility: Conversations still make sense for QA team
• ✅ Consistency: Same customer = same token everywhere
• ✅ Scale: Process thousands of calls in minutes
• ✅ Compliance: GDPR, CCPA, PCI-DSS all covered

🏆 Your boss is happy. Legal is happy. Customers are protected. And you didn't even need to work on Saturday.