Skip to main content

MaestroQA PII Redaction

Domenic Nucci avatar
Written by Domenic Nucci
Updated yesterday

MaestroQA Redaction is a feature available in-application for self-service setup. When enabled, this feature hides sensitive information in call recordings, transcriptions, and digital comments (public and private) in MaestroQA.

What is Redacted?

  1. Transcriptions + Digital Comments

    1. PII words flagged within the transcription / digital comments will be redacted

  2. Call Audio

    1. PII words detected within the audio player will be silenced

Call Redaction (Transcription + Call Audio) is enabled separately from Digital Comment redaction. That means you can redact calls but not digital comments, or vice versa. If both are enabled, they will redact the same entities outlined below.

When is Redaction Run and what is Stored?

Unredacted calls and/or digital conversations are temporarily stored in a logically and network separated area of MaestroQA's infrastructure so MaestroQA can run it's redaction process. Once the redaction process is run, the redacted content is moved into the permanent storage component of MaestroQA's infrastructure where it is made visible to end users of MaestroQA.

The temporary storage that contains the unredacted content is fully deleted on an hourly basis.

How to Configure Redaction

The system administrator can go to the Settings > Transcription Settings page to enable this feature

You can configure which types of PII are redacted in the selector. It provides a default set of options but you can customize it to redact any of the types listed here: https://microsoft.github.io/presidio/supported_entities/

Accuracy

MaestroQA redacts information using probabilistic methods, combining pattern matching, context analysis, and checksum validation. Pattern matching identifies data like phone numbers or credit cards based on known formats, while context analysis uses surrounding keywords to improve accuracy. Checksum validation ensures values like credit card numbers are likely real, reducing false positives.

Because detection is probabilistic, redaction is not guaranteed to be perfect and may occasionally miss or misidentify content

Did this answer your question?