People will only speak honestly if they understand why they are being asked, how their input will be used, and what will not happen to them as a result. Voxant is designed with UK GDPR principles, recognised AI trust guidance, privacy-aware reporting and human oversight in mind.
Voxant is currently in private beta. The controls described on this page apply to the private beta unless a capability is explicitly identified as in development or subject to client setup.
Every protection below is a mechanism built into the product, not a policy promise.
Everyone in scope is interviewed and every voice feeds the synthesis, including teams of fewer than five. What a group that small never gets is its own breakout line or a quote published under its label, because either would identify the people in it. If only four people in a department take part, they are still heard and their answers still count towards the whole-organisation findings.
Verbatim quotes carry at most a role level and a department label, and a label appears only when that group passes the floor of five. A quote is never attributed to a name, and never to a group small enough to identify anyone.
No manager sees who spoke, what was said, or who declined. Raw transcripts and interviewee identity reach only a consultant engaged in overwatch on that engagement; every other viewer, including owners and executives, receives a redacted view with no names attached.
Whether any individual completed or declined an interview is never reported to the employer.
The employer sees aggregate synthesis with de-identified verbatim quotes. Transcripts stay out of employer hands.
Voxant supports organisational understanding. It does not make employment decisions, rank individuals, infer emotion, or replace human judgement.
Before the interview, people are told they are speaking with an AI system, why the interview is being run, and how the output will be used. The interviewer discloses its AI identity at the start of every conversation, and people can pause, skip a question or stop at any point.
Interview evidence, readiness signals, workflow friction, confidence gaps, governance themes and practical blockers.
It does not rank employees, infer emotion, automate HR decisions or replace human judgement.
Reports carry aggregate organisational insight, with evidence separated from inference and recommendation.
HRIS data is used for cohort design, segmentation and invitation routing. Direct identifiers are held in an encrypted identity vault, separate from analysis data, and operational records use pseudonymous person codes.
Reports support consultant and leadership judgement; they do not bypass it. A consultant can be engaged in overwatch at every stage.
Voxant is not certified against these frameworks; it is designed with them in mind. Each is used as a reference for AI risk, privacy and governance practice.
Used as a reference point for trustworthy AI characteristics such as validity, safety, security, accountability, transparency, explainability, privacy enhancement and bias management.
Used as a UK-facing reference for accountability, transparency, lawfulness, fairness, accuracy, security and data minimisation in AI systems.
Used as a policy-level reference for human-centred, trustworthy AI that respects rights, democratic values, accountability and responsible stewardship.
Voxant runs on a small number of vetted, enterprise-grade providers, each operating under its own data processing agreement and UK GDPR-aligned terms. The functions they perform are set out below. The current named sub-processor list, with each provider's data processing terms, is provided to clients and prospective clients on request under NDA as part of the trust and security pack.
Connection and usage data
Interview transcripts, evidence, reports and account records
Live interview audio and transcripts
Interview transcript text (no audio)
Recipient names and email addresses
Connection signals such as IP address
Interview transcripts and interviewee personal data are anonymised once they pass the retention window, which is configurable per client and set out in the trust and security pack. A scheduled job runs daily: it clears the raw transcript, removes names and email addresses, and deletes the encrypted identity records of anyone whose interviews have all aged out.
Report generation reads only de-identified aggregate fields and the synthesised report, never the transcript, name or email. Deleting personal data does not delete the organisation's deliverable.
Anyone interviewed can request erasure of their data at any time. Erasure clears that person's transcripts and identifiers immediately, regardless of the retention window.
Each purge and each erasure is recorded as a structured retention event and an audit log entry, so the deletion history is itself auditable.
Live interview audio is processed by our voice AI sub-processor solely to run the live conversation and produce the transcript. The provider, its audio retention and its processing region are set out in the sub-processor pack available on request.
The processing region for each sub-processor is documented in the sub-processor pack, provided to clients and prospective clients on request.
Everything above describes the controls currently implemented in the Voxant private beta unless a capability is explicitly identified as in development or subject to client setup. The items below are in development and are labelled as such wherever they appear.
Organisation data is imported today through a secure route that separates direct identifiers from analysis data on entry. Direct certified connectors to HRIS platforms are in development, subject to client setup, access approval, security review and data processing arrangements.
Plain answers. Voxant is for organisational understanding, not individual judgement.
No. Voxant is designed to identify organisational patterns, never individual behaviour. It does not rank employees, infer emotion, make employment decisions or replace human judgement.
Yes. The interviewer discloses that it is an AI interviewer at the start of every conversation, and interviewees are told the purpose of the interview and how their responses will be used before it begins.
No. Reports are aggregate: no group smaller than five gets its own breakout line and no quote is published that could identify one, though those answers still count towards the whole-organisation findings. Raw transcripts and interviewee identity reach only a consultant engaged in overwatch. Managers and executives see synthesis, never transcripts.
HR and organisation data is used to map cohorts, send invitations and structure reports. It is not used to score individuals or to automate employment decisions.
Organisation data is imported through a secure route that separates direct identifiers from analysis data. Certified connectors to HRIS platforms are in development, subject to client setup, access approval, security review and data processing arrangements.
No. Voxant supports organisational insight and human judgement. It is not designed for hiring, promotion, redundancy, performance management or disciplinary decisions.
That is part of the value. Voxant is designed to surface hidden risks, resistance, confusion and bottlenecks before they become failed pilots, cultural damage or performance problems.
A trust and security pack is available on request, covering data flow, retention, access controls, integration scope, privacy language and non-decisioning policies.