4,830 candidate prequalifications completed in three weeks, with data processed in Europe

Hippolyte.ai needed a voice natural enough to keep candidates engaged in the conversation, while meeting the compliance requirements of its key accounts. We managed to reconcile both, with no compromise.

4,830 voice screening interviews  –  272 hours of live conversation  –  nearly 20,000 interviews projected per quarter

Hippolyte.ai is a French recruitment platform. Its prequalification solution works in four steps: it measures how well CVs match job postings, conducts screening interviews in writing or by voice (Screen.ia module), assesses candidates' soft skills, then gives the recruiter a shortlist of profiles ranked by relevance, on which the recruiter makes the decision.

Its model rests on two essential conditions: giving candidates an experience smooth and natural enough that they complete the interview, and guaranteeing key accounts a level of data protection that meets their requirements.

ElevenLabs steps in on Screen.ai's voice channel, precisely where these two challenges meet.

The challenge: a prequalification interview that is smooth for the candidate, and complete and usable for the recruiter 

Hippolyte.ai first tested a generic text-to-speech solution for its interviews. But the voice lacked naturalness, latency was noticeable and the exchanges lacked fluidity.

 In recruitment, these details directly affect the candidate experience. An artificial or unpleasant conversation, and the candidate hangs up: the company loses a profile it spent time and money sourcing

The second constraint was just as important commercially. Hippolyte.ai works with key accounts that closely examine the conditions under which candidate data is processed before signing. Any provider involving data processing outside the European Union could therefore become a blocking point.

“Our clients are large enterprises, and they audit us line by line on candidate data protection. A voice provider outside the European Union would have closed the door on those contracts.”

Gauthier Bailleul — CEO, YES WE CODE, the company behind Hippolyte.ai

Why Hippolyte.ai chose ElevenLabs

ElevenLabs naturally stood out in Hippolyte.ai's research as a reference in speech synthesis. But the final decision came down to three concrete criteria:

• Voice quality validated in real interview conditions, with a contextualized agent that adapts the exchange to the role and to the candidate's CV, previously anonymized and analyzed by Hippolyte.ai's engine, and not only from a demo script.
• Data hosted and processed within the European Union. Hippolyte.ai hosts its own data in France with OVHcloud and did not want to compromise on this point.
• A Data Processing Agreement (DPA) contractually governing the purpose of processing and the data retention period, in order to meet in particular the compliance requirements and security questionnaires of its key accounts.

With these three criteria met, ElevenLabs became the voice channel provider for Hippolyte.ai, under a DPA that notably provides for automatic deletion of raw audio recordings within 24 hours of their transcription.

An integration entirely carried out by the Hippolyte.ai team

ElevenLabs provides speech synthesis and call management for one specific step of the journey: the voice interview within Screen.ai. Everything else is designed and operated directly by Hippolyte.ai.

• Applications arrive via the Hippolyte.ai API, through file import or by syncing with the client's ATS.
• CVs are anonymized, vectorized, then scored for relevance against the job posting.
• Candidates above the defined threshold receive an invitation by SMS or email, along with a secure link.
• They can choose between a questionnaire, a written chat or a voice interview. ElevenLabs operates on this last channel.
• The transcript is anonymized, then analyzed by Hippolyte.ai's self-hosted open-weights model, on dedicated GPUs in France, without using a third-party AI API.
• The recruiter then receives a score and a summary. The final decision always remains human.

Hippolyte.ai's technical team built and configured the entire integration without any involvement from our field engineering team.

For a platform that needs to roll out new features quickly to its own customers, this is an important point: putting the voice channel into production required no additional professional services.

Key criteria for meeting key account requirements

Here is what Hippolyte.ai can present to the security and legal teams of its key accounts:

• Candidate data is hosted and processed in France. Voice processing by ElevenLabs takes place within the European Union.
• The open-weights language model is self-hosted on dedicated GPUs: candidate data does not pass through a third-party AI API for semantic analysis.
• Data that could identify candidates is systematically anonymized before AI processing.
• No decision is fully automated: the system prioritizes and summarizes information, but the decision rests with the recruiter.
• A daily human review is carried out by sampling the analyses produced by the AI.
• No biometric data is collected via the voice channel: voice is used only for transcription, never for identification.

“A voice screening interview leaves no room for approximation. If the voice sounds off, the candidate hangs up. The quality of the rendering changed how candidates experience the exercise.”

Lucie Postel — Product Manager, Hippolyte.ai

4,830 interviews in three weeks, with no compromise on data residency

Over the first three weeks of September 2026, Hippolyte.ai conducted 4,830 voice screening interviews, representing 272 hours of conversation.

The average interview lasted 3 minutes and 23 seconds, with a peak of 640 calls in a single day, versus an average of 302 calls per business day over the period.

At this pace, the voice channel should represent nearly 20,000 conversations and 1,125 hours per quarter. And these volumes continue to grow as Hippolyte.ai activates the module with new clients.

Above all, this ramp-up was achieved with no compromise on the data residency and processing requirements that had been set from the start.

And now?

The first growth driver is volume: more clients are activating the voice module, while existing clients are gradually increasing their usage.

The second is language. Hippolyte.ai plans to extend its industry dictionaries from French to English and Spanish, which would eventually enable multilingual voice screening interviews.

Hippolyte.ai's goal was to demonstrate that it was possible to use conversational AI in recruitment without sacrificing either the candidate experience or data protection requirements.

Over the first three weeks of September, the first results show that it is possible to combine voice quality, scaling and compliance requirements.