The increasing use of AI in customer service introduces significant risks around the handling of sensitive information such as Social Security Numbers, credit card details, and medical identifiers. Traditional text-based redaction techniques, such as regex or NER, struggle with real-world challenges, including transcription errors, accents, and contextual ambiguity. This paper proposes a multimodal framework for automated personally identifiable information (PII) detection and redaction in voice transcripts. The framework integrates acoustic signals, speech-to-text transcripts, and contextual metadata to improve reliability in identifying sensitive data. A pipeline architecture is outlined where audio features complement textual cues, enabling real-time detection and redaction during streaming analytics. The methodology emphasizes conceptual design, qualitative analysis, and case-based reasoning, supplemented by a small-scale proof-of-concept validation on synthetic data demonstrating improved recall over text-only baselines. Illustrative scenarios in healthcare and financial services demonstrate the framework’s compliance benefits, highlighting its ability to reduce regulatory risk and strengthen customer trust. The proposed design supports enterprise-scale deployment.