We have updated our Speech Analytics offering in 2025; here are the key points to keep in mind.
Speech Analytics Offering at a Glance
The service involves analyzing audio recordings and transcribing their content into text using Speech-to-Text (STT), or “automatic speech recognition” in French, an assistive technology capable of automatically converting spoken content into written text.
The goal is twofold:
- Analyze customer feedback to quantify and understand the reasons for contact. Conversation analysis will expand the volume of feedback by enabling the analysis of all recorded calls, thereby providing a larger dataset to analyze compared to survey responses, for example.
- Analyze the Advisor's Voice to automatically assess call compliance.
How does it work?
We use Speech-to-Text (STT) technology to transcribe audio into text and a Large Language Model (LLM) to identify the topics discussed by clients and automatically populate the compliance checklists you use for your projects.
The benefits
The benefits are as follows:
- Significant time savings on call monitoring: You no longer need to manually listen to every call, freeing up valuable time to coach agents—implementing action plans, conducting training sessions, reviewing procedures, and more…
- Enhanced reliability: 100% of recordings are analyzed, without human bias
- Better quantification of contact reasons: Data from voice servers and agents’ manual coding entries rarely allow for a reliable quantification of contact volumes by reason; STT provides an unbiased source to help you gain clarity and focus on the reasons why your customers contact your Customer Service department
- Identifying customer pain points to guide corrective actions
- Identifying risky situations (non-conformities, unacceptable situations)
- A better understanding of ambiguous situations: the gap between customer perception and compliance
Key Points on Activation
The service can be technically launched within a few weeks; however, finalizing the legal details may take longer.
Here are the main steps for implementation:
- Retrieval of audio files via automatic file upload for real-time processing
- Creating the account and user accounts
- Configuring the compliance grid(s) (sections, items, weightings)
- the definition of the semantic classification scheme
- Data integration (including any necessary transformations)
Once the system is up and running, a calibration phase will be necessary to tailor the prompts to the quality teams’ existing practices and gain their buy-in.
Legal considerations
The main legal issues that need to be addressed are as follows:
- Any recording of a conversation requires prior notification to customers, who must give their consent
- Determine the volume of calls to be recorded based on the defined purpose: When the purpose of the recordings is for quality assurance or training, the CNIL recommends a recording rate of no more than 30%
- Set a limited retention period that is consistent with the defined purposes of processing
- Restrict access to recordings to authorized personnel
- Grant customers the right to access and object to the processing of their personal data upon request
