Redaction Policies
ASAPP employs a default policy for all implementations. For each implementation, a policy can be further customized to include or exclude certain types of redaction to suit the targeted use case(s).
Some redaction rules are always required to be implemented. Others are required to be implemented in situations where agents are taking payments directly.
Required policies must be used unless both ASAPP and the Customer’s security assurance leaders (CISO equivalent or delegate) approve that they are not necessary for the scope of the implementation.
Default Policy
All rules in the default policy are applied immediately to (1) utterances from all participants in a conversation and (2) all agent-written summary notes.
This default policy applies to training data sent to ASAPP and operational data used by and stored as a result of live implementations.
| Rule | Requirement |
|---|---|
| Credit Card Numbers | Always Required |
| Social Security Numbers | Always Required |
| 9-Digit Numerics | Always Required |
| 3-Digit Numerics | Default* |
| 4-Digit Numerics | Default* |
| 10-Digit Numerics | Default |
| E-mail Addresses | Default** |
| Phone Numbers | Default |
| Voice Credit Card Numbers | Default |
*Required in situations where agents are taking payments directly. **Email addresses rule is a default for digital messaging conversations only. For more information, visit email redaction capabilities.
By default, ASAPP also applies redundant credit card number and social security number redaction to all stored operational event data prior to persisting it. This is not configurable.
AI Services Redaction
ASAPP applies redaction policies to the following types of data:
- Data provided to ASAPP for initial model training prior to implementation
- Data ingested or outputted as part of a service’s regular operation
Redacted data elements are outlined below for each product. Reach out to your ASAPP account contact to configure a redaction ruleset suited to your use case.
AutoCompose
Redacted Training Data:
- Historical conversation transcripts provided to ASAPP for generating an agent global response list specific to a given implementation
Redacted Operational Data:
- Conversation transcript messages provided to ASAPP
Read more about AutoCompose here.
AutoSummary
Redacted Training Data:
- Historical conversation transcripts and summary notes provided to ASAPP for training summarization models and for generating an intent hierarchy specific to a given implementation Redacted Operational Data:
- Conversation transcript messages provided to ASAPP
Read more about AutoSummary here.
AutoTranscribe
Redacted Training Data:
- Recorded call audio files provided to ASAPP for speech-to-text model training
- Written transcripts generated from historical call audio files provided to ASAPP for speech-to-text model training
For training audio files only, redacted portions are removed from the audio itself using the default redaction policy. In these training test sets, audio is redacted with a beep sound or discarded entirely.
Redacted Operational Data:
- Transcript outputs generated from supplied audio streams
In addition to the default redaction policy, transcript outputs are also subject to context-aware redaction rules to detect parts of conversations where credit cards and SSNs are likely being discussed.
Read more about AutoTranscribe here.
Applications Redaction
ASAPP applies redaction policies to the following types of data:
- Data provided to ASAPP for initial model training prior to implementation
- Data ingested or outputted as part of a service’s regular operation
Redacted data elements are outlined below for each product. Reach out to your ASAPP account contact to configure a redaction ruleset suited to your use case.
Messaging Application
Redacted Training Data:
- Historical conversation transcripts provided to ASAPP for generating an agent global response list specific to a given implementation
- Historical conversation transcripts provided to ASAPP for generating an intent hierarchy specific to a given implementation
- Historical conversation transcripts and summary notes provided to ASAPP for training summarization models and for generating an intent hierarchy specific to a given implementation
Redacted Operational Data:
- Conversation messages submitted by customers and agents
- Summary notes submitted by agents
Voice Application
Redacted Training Data:
- Recorded call audio files provided to ASAPP for speech-to-text model training
- Written transcripts generated from historical call audio files provided to ASAPP for speech-to-text model training
- Historical conversation transcripts provided to ASAPP for generating an intent hierarchy specific to a given implementation
- Historical conversation transcripts and summary notes provided to ASAPP for training summarization models and for generating an intent hierarchy specific to a given implementation
For training audio files only, redacted portions are removed from the audio itself using the default redaction policy. In these training test sets, audio is redacted with a beep sound or discarded entirely.
Redacted Operational Data:
- Conversation utterances transcribed by ASAPP
- Summary notes submitted by agents
In addition to the default redaction policy, transcript outputs are also subject to context-aware redaction rules to detect parts of conversations where credit cards and SSNs are likely being discussed.