Why a leading APAC airline chose Level AI to modernize QA and protect passenger data

A tier-one international carrier replaced a legacy quality system that scored less than two percent of customer interactions, leaving the operations team with unvalidated performance data. Partnering with Level AI, the airline implemented dual-layer redaction to protect sensitive passenger files, established automated quality scoring across all voice and digital channels, and analyzed customer insights derived directly from conversation analysis.

• The Customer: A premier Asia-Pacific airline operating a global network, employing more than 3,500 contact center agents who handle upward of 1.8 million customer interactions every month. Operations run day-to-day through a global business process outsourcing (BPO) partner and rely on Cisco for primary voice recording.
  
  

• The Challenge: The airline's legacy quality assurance platform, Verint, failed to integrate with core recording infrastructure, resulting in significant reporting gaps. More critically, the airline's newly formed AI Safety Council and information security team placed a strict embargo on vendor evaluations. They refused to share passenger data or authorize a proof of concept until a vendor proved it could isolate and redact sensitive information before processing.
  
  

• Why They Chose Level AI: Google had the inside track, but the airline chose Level AI based on relationships, trust, and proven technical capabilities. Level AI satisfied rigorous security mandates through dual-layer redaction for both audio and text. Additionally, Level AI provided aviation-specific language models that achieved 100% accuracy in scoring during the 50,000-interaction pilot, surpassing Verint’s legacy keyword-based systems.

The Challenge: Dual-Layer Evaluation and Legacy Limits

The airline managed its high-volume contact center network through a strategic BPO partner. While the BPO partner held operational responsibility and coordinated the technology evaluation, the airline retained absolute veto power over any software entering their environment.

The existing quality system, Verint, failed to deliver reliable performance metrics. Poor data integration between the quality platform and Cisco call recorders resulted in persistent coverage gaps, leaving leadership with no clear visibility into customer sentiment or agent compliance.

"We were blind to over 98% of our customer interactions because legacy sampling only gave us a random handful of scores. Moving to a partner that handles the volume while meeting our security mandates changed our entire operations strategy."

— Quality Assurance Director, Global BPO Partner

The Security Gate: Why Text Redaction Was Not Enough

For highly regulated transit brands, data privacy is a binary requirement. The airline's internal AI Safety Council demanded absolute assurance that no personally identifiable information (PII) or payment card industry (PCI) data would leave their secure partition, even for a trial or proof of concept, without verifiable proof that sensitive passenger details would be completely removed.

This requirement presented a significant technical barrier. Standard industry solutions typically redact text transcripts but leave the underlying raw audio files untouched. Because agent conversations frequently involve spoken passport details, reservation codes, and credit card numbers, the unredacted voice recordings remained a major compliance risk.

The airline required a platform capable of dual-layer redaction:

• Audio-Wave Redaction: The system must locate and actively strip sensitive spoken numbers directly from the raw audio files, rendering the audio clips safe for storage and review.

• Transcript Redaction: The corresponding text transcripts must be sanitized simultaneously, removing names, loyalty numbers, and payment details.

Until a vendor could demonstrate this dual-layer capability, the AI Safety Council refused to schedule technical discovery calls or permit the BPO partner to advance the RFP.

"Our brand is built on passenger trust. If we cannot guarantee that a credit card number or passport digit is scrubbed from the actual audio wave before it leaves our partition, the technology simply does not enter our stack."

— Director of Information Security & AI Governance, APAC Airline

The Need: Complete Coverage, Customer Insight, and Custom Security Controls

To support a wider customer experience initiative, the quality and customer experience (CX) teams outlined specific operational requirements for the new platform:

• Operational Validation: Pass a mandatory 50,000-interaction proof of concept (POC) overseen by both the BPO partner and the airline's IT department.

• Comprehensive Quality Scoring: Transition from small-sample audits to automated scoring across 100 percent of voice and digital interactions to eliminate reviewer bias.

• Intent Analysis: Evaluate the context of customer conversations beyond simple keyword matches, distinguishing actual customer issues from coincidental word choices.

• Direct Customer Insights: Convert unstructured interaction data into structured feedback that business analysts can query directly, bypassing slow and low-response post-call surveys.

• Aviation-Specific Understanding: Process complex routing terms, booking codes, and frequent flyer nomenclature accurately without requiring massive general-purpose computing resources.

Why Level AI: Deep Redaction and Domain-Specific Intelligence

Despite Google’s early presence and local leverage, Level AI was selected as the partner of choice following a rigorous evaluation. The carrier selected Level AI based on established trust and a demonstrated ability to satisfy every technical requirement, a decision made by three foundational architectural advantages:

Dual-Layer Audio and Transcript Redaction

Level AI resolved the security council's central objection by deploying automated redaction across both written transcripts and raw audio files. The platform identifies sensitive data fields, including unique airline loyalty formats, custom booking reference numbers, and passport structures, and redacts them from the audio file and text transcript simultaneously. This capability satisfied the compliance team's requirements and allowed the safety council to approve the 50,000-interaction proof of concept.

Domain-Specific Small Language Models (SLMs)

Instead of relying on generic, off-the-shelf public models, Level AI demonstrated specialized AI Core architecture. The platform employs domain-specific Small Language Models (SLMs) trained on aviation terminology, travel lexicons, and the airline's unique scoring guidelines.

This targeted approach provides highly accurate intent and sentiment analysis while operating within a secure, dedicated environment. The models distinguish between distinct intents, recognizing the difference between "I need to cancel my booking" and "I am not looking to cancel my booking," which legacy keyword tools often flagged identically.

"General-purpose AI models struggle with the unique vocabulary of the airline industry—codes, booking systems, and complex routing rules. Training the system on our actual interaction data gave us the accuracy our operations required from day one."

— CX Technology Lead, APAC Airline

Automated Quality Assurance at Scale

The QA team lead was impressed by Level AI’s QA-GPT, which scored nearly 100 percent of incoming calls, chats, and emails against existing evaluation criteria during the POC. This automated coverage identifies compliance issues and training opportunities across the entire conversation, rather than relying on a tiny fraction of audited calls.

The Path Forward: Scaling Compliance-First QA

Following the successful completion of the 50,000-interaction proof of concept, the airline cleared all outstanding security requirements. Operational teams are now preparing to fully transition from their legacy system and integrate Level AI with their global Cisco recording architecture.

By resolving the security gate first, the airline's BPO partner can now monitor agent performance and extract customer insights across all channels. The carrier is positioned to improve reservation accuracy and reduce handling times while maintaining complete compliance with global data privacy standards.