Conversations: workplace roleplay with transcript-backed evidence

What it is for

Conversations is a voice-based workplace roleplay. The AI plays someone the candidate might need to speak to, such as a client, colleague or subordinate. It automates the work of running that interaction and recording evidence for recruiter review.

A job description or brief can generate the roleplayer, task and suggested behaviours. The recruiter can refine every element with the operator’s assistance. The assessment is concerned with behaviours shown in the conversation, rather than whether the candidate recites a particular answer.

What candidates experience

Candidates speak with the AI roleplayer in a fully audio conversation. They can ask questions, clarify the situation and work towards the stated task. The roleplayer’s background and concerns give the interaction a workplace context.

Candidates receive their assessment invitation by email and complete the configured stages within the campaign’s deadline. They can email for technical help. The recruiter decides any assessment exception; support does not change the scenario for an individual candidate.

What recruiters receive

Recruiters receive a report and the transcript. Each behaviour finding explains whether the behaviour was fully or partially demonstrated and why. The report links to relevant transcript passages and playable audio clips so the reviewer can inspect the evidence directly.

The AI judgement is deliberately narrower than “is this person suitable?” It considers the predefined behaviour in the particular transcript. Where a score is shown, it combines those findings using fixed rules and recruiter-adjustable weights. The score is guidance, not a validated prediction of job performance.

Worked example: clarifying a stakeholder request

Illustrative example: an internal client asks a legal candidate for an urgent contract change without fully explaining the commercial priority. The candidate can clarify the objective, identify constraints and agree the next step with the roleplayer.

A report might identify the exchanges where the candidate checked the client’s requirements and explained a trade-off. The reviewer can read those passages and play the associated clips. Several approaches may demonstrate the same behaviour; the assessment is not a test of guessing one exact wording.

Reviewing and correcting findings

A recruiter can disagree with an interpretation and correct findings in the UI. That correction can affect the individual report. The source exchange should make it possible to explain the disagreement against the agreed criteria.

A sufficient body of labelled evidence and later outcomes may support future improvements to the assessment. Outcome-linked predictive analysis is future work, rather than an established prediction available from today’s conversation report.

Limitations and human review

A transcript can contain errors, and a short roleplay samples only part of someone’s communication and judgement. A deterministic calculation does not make the AI’s underlying interpretation infallible. Review the relevant exchange when a finding matters to a decision.

Decision Agent does not infer emotion or intention from voice. Its current assessment claim concerns predefined behaviours in responses. The hiring team remains responsible for the process, the interpretation of the evidence and any exceptions.

Explore the interactive assessment and report examples, read about human review in an AI-assisted pipeline, or see module credit costs.