Objectives of the service
The incoming and outgoing calls of the brands can be recorded by the call processing and/or distribution tools. Thus, the brand has a large amount of data that can be exploited for different purposes.
To be able to process all this data efficiently, the audio recording, even if accompanied by metadata specifying the context of the call (date, agent, call queue, competence, etc.), does not have the capacity to be exploited efficiently: the visibility of its content, the themes addressed during the call, the purposes of the exchange, the satisfaction of the customer, etc. are all pieces of context that would require listening to each call in detail in order to be annotated.
However, having the capacity to quickly measure trends, understand elements of blockage or frustration, whether from the customer or the agent, and capture the signals marking shortcomings in a brand's processes is a strong challenge, requiring mass processing skills.
In order to process these recordings and improve understanding of the content contained in these calls, Allo-Media offers a service called "Batch Transcription", enabling any brand to transcribe their audio recordings and to obtain an accurate transcription in return. To do this, Allo-Media uses a proprietary natural language speech recognition technology, ensuring a transcription adapted to the specific technical constraints of the telephone environment: degraded sound quality, spontaneous speech, speech breaks, various regional, national and international accents, etc...
In order to further improve the usability of this data, Allo-Media also offers to add enrichment elements to the transcription, both to secure confidential or sensitive data and to display analytics elements. Thus, Allo-Media processes personal identifiable information in real time, and is able to display a label instead of the actual transcription ("First name" instead of "Nathan"; "City" instead of "Liverpool", details of the possibilities are described in the technical documentation). Similarly, Allo-Media offers several elements for analyzing the conversation, in the form of tags, offering a readability of the elements making up the call, without having to listen to it in full:
- tags specific to the customer's voice: satisfaction, effort, intentions, etc.
- tags specific to the advisor's voice: quality monitoring, compliance, etc.
- tags dedicated to understanding interactions and purposes: conversions, products, etc.
Details and clarifications
Overview of the processing chain
Data input protocols
To enable transcription of the audios pre-recorded by the brand, Allo-Media offers an exchange protocol with one of the following entry point :
- an SFTP server, allowing audios to be uploaded ready to be processed by Allo-Media. This SFTP server is hosted in the Allo-Media Cloud.
- For any access request, please send your request to: email@example.com
- The connection modalities will be specified in return of this access request.
- The data entrusted by the brand will then be processed, as described in the following paragraph, and will then be permanently deleted from the SFTP server.
- All technical details related to this input protocol are available in the Allo-Media technical documentation.
- an integration to Axialys' API, allowing available audios to be automatically fetched
- All technical details related to this integration are available in the Allo-Media technical documentation.
Processing carried out
As soon as data is made available on the Allo-Media platform, an automated process is initiated.
Depending on the processing options chosen, the audios filed by the brand are subject to :
- a transcription (mandatory), in the language chosen and defined within the processing configuration (details in the technical documentation)
- an enrichment by redaction of personal/sensitive data (option) (details of the possibilities described in the technical documentation)
- Enrichment with speech-analytics elements (optional, and depending on the certification)
Data output protocols
Once the processing by the Allo-Media infrastructure is complete, the now transcribed, structured and enriched audios can be consumed in different ways:
- via the Scribr user interface (optional, and depending on the certification), displaying the calls in the form of a log, with various structured elements: audio of the call, transcription, enrichment metadata and offering export possibilities of this log. (details of the possibilities are described in the technical documentation)
- via the Activate API (option), exposing all call data and metadata (details of possibilities described in the technical documentation)
- via integrations with third-party tools, offered off-the-shelf by Allo-Media (option), allowing Allo-Media's data model to be transformed into the third-party recipient's data model (details of possibilities described in the technical documentation)
Since Allo-Media acts as a service provider, it is necessary to establish with the brand, which is solely responsible for data processing, a configuration adapted to each of the processes carried out. Thus, the following elements can be configured, affecting the availability of data within the Allo-Media servers
- the retention period for the audios (in increments of 1 day - default value recommended by the CNIL: 180 days)
- the retention period for transcripts (in increments of 1 day - default value recommended by the CNIL: 360 days)
- Named entities requiring redaction