Voice AI for HR Screening

When a candidate applies, the system sends an SMS or email with a link to schedule the AI phone screen — or calls them directly if your process calls for immediate outreach. The AI conducts a structured interview: it asks your configured screening questions, follows up on answers that need clarification, handles interruptions and off-topic responses naturally, and adapts the conversation flow based on the role type. After the call, a transcription pipeline runs (Deepgram), responses are scored against your rubric (GPT-4o evaluation), and a ranked summary is written to the ATS. The recruiter sees a scored candidate profile with the key quotes from the call, not a raw recording. The process takes the same time as a human screen — 15–20 minutes per candidate — but runs at unlimited concurrency, any time of day.

Yes. Question sets and scoring rubrics are configured per role type. A retail associate screen asks different questions and weights different attributes than a software engineer screen or a warehouse operative screen. We build the question library and scoring logic in collaboration with your recruiting team, typically in a 2-week configuration phase before the system goes live. The configuration is managed through an admin interface — your recruiting team can adjust questions, add new role types, and update scoring criteria without engineering involvement.

Disclosure is mandatory and built in — the system identifies itself as an AI screening tool at the start of the call. In our deployments, drop-off rates from disclosure are lower than drop-off from scheduling delays (the average time-to-first-contact for manual screening is 3–5 business days; AI screening contacts within hours of application). Candidates who complete the screen and move forward report the experience positively because it's fast, structured, and available on their schedule — 11pm screens are common. Candidates who weren't going to progress typically don't object to the format.

We integrate natively with Lever, Greenhouse, and Workday Recruiting for candidate profile updates, stage progression, and summary note writing. For Workday: we use the Candidate REST API to write scored evaluation notes and progress candidates between stages. For Lever and Greenhouse: we use their webhook + API combination to trigger screens on new application events and write feedback at completion. For custom or less common ATS platforms, we build the integration. We also integrate with LinkedIn Talent Hub, SmartRecruiters, and iCIMS on request.

The system is built with compliance as a design constraint, not an afterthought. GDPR: candidates are informed that the call will be processed by AI, transcribed, and the summary stored. Data retention periods are configurable per your policy — transcriptions and summaries are deleted on schedule. TCPA (US): outbound call timing restrictions, consent verification, and Do Not Call registry checks applied automatically. EEOC (US): scoring rubrics are reviewed to avoid questions that would capture protected class information; the evaluation criteria are documented for audit purposes. We produce the compliance documentation your legal team needs to review the system before go-live.

A focused build for a single role type — question set, scoring rubric, one ATS integration, and the call infrastructure — typically runs $25,000--$40,000 and takes 6--10 weeks. A multi-role-type platform with a question library, recruiter admin dashboard, configurable rubrics, and multiple ATS integrations typically runs $45,000--$75,000. Per-call infrastructure costs (Twilio, Deepgram, GPT-4o) run approximately $0.30--$0.80 per completed screen depending on call length and the number of follow-up evaluations — substantially less than the all-in cost of a recruiter hour.