Amazon Connect Talent ties every candidate score back to a line in the transcript

Two lines in the launch documentation matter more than the feature list. AWS says each competency is scored against a rubric that defines what a strong response and a weak response look like, and that every score is tied to specific evidence from the interview, so the result traces back to what the candidate said ^[7]. Recruiters see that reasoning beside the score, with full transcripts, in a dashboard ^[16]. The evidence tie is good engineering, and it is the part of the product an audit can actually use.

The GA capability list includes AI-led voice interviews with adaptive questioning ^[3]. The launch blog describes a consistent, structured interview ^[15] and says organizations get the same standard of evaluation whether they are evaluating the first candidate or the thousandth ^[13]. Both hold when consistency is a property of the rubric. Two applicants for the same job can be asked different follow-ups and still be scored against one definition of a strong answer.

AWS says all candidate data is anonymized during AI evaluation, removing factors that can introduce unconscious human preconceptions ^[8], and that the assessments are designed around competencies associated with success in the role, not subjective impressions like confidence level or speaking style ^[9]. Candidates are told what data is collected, how it is used, what is not collected, and what to expect before they proceed ^[17]. The interviews are voice interviews ^[2]. Whether the scoring model reads the audio or a transcript of the audio decides what there is left to anonymize. Neither post describes how anonymization applies to audio, and AWS did not publish a price.

Both posts say the service is informed by decades of Amazon's hiring science ^[11], and the blog notes that Amazon is one of the world's largest employers ^[12]. It is a benchmark drawn from someone else's workload: competencies calibrated against Amazon's roles, at Amazon's application volume. For it to transfer, your definition of a strong answer for a warehouse shift lead has to resemble theirs. Recruiters configure the evaluation criteria, the assessments and the interview questions themselves ^[6], so the calibration work lands on whoever writes the criteria.

The GA post calls the output "detailed candidate evaluations generated by their AI teammate" ^[20]. The blog says recruiters keep final decision authority over every hire ^[14], and the list those recruiters decide from is the one the model ordered.