Automate Recruiting Follow-up with Custom Voice AI

Recruiting teams often try using their ATS's built-in email automation but find open rates are too low for time-sensitive roles. They turn to voice but hit a wall with off-the-shelf tools. Call-blasting services like CallFire are just dumb dialers. They play a pre-recorded message but cannot understand or react to a candidate's response. When a candidate says, “I’m busy, call me back,” the system has no way to process that information.

A 10-person recruiting firm tried one of these services to follow up on 200 recent applications. The message said, “Press 1 to speak to a recruiter.” Over 90% of candidates hung up immediately. The few who pressed 1 were dropped into a general queue, creating a confusing experience for both the candidate and the recruiter who had no context on the call. The tool could not answer simple questions or schedule a callback, resulting in a wasted budget and annoyed applicants.

Even modern recruiting CRMs with so-called AI features fall short. Their voice tools rely on low-quality, generic text-to-speech engines that make the outreach feel impersonal and cheap. Customization is often limited to inserting a candidate's name into a rigid template. You cannot change the voice, alter the intonation, or build logic to handle a simple question, making the interaction feel robotic and ineffective.

Syntora would approach building a custom voice AI system by first conducting a detailed discovery to understand your existing Applicant Tracking System (ATS) and define specific trigger events. This includes identifying necessary candidate data such as name, phone number, and job applied for. The initial architectural design would typically involve setting up a Supabase table for defining and adjusting triggers, allowing non-technical users to manage timing without code changes.

The core of the system would be a Python application, designed for deployment on AWS Lambda or similar serverless infrastructure, enabling cost-effective scaling. When a trigger condition is met, this function would execute. It would integrate with a text-to-speech API like ElevenLabs to generate natural-sounding voice prompts. The conversational flow would be dynamic, for example: "Hi, [Candidate Name], this is [AI Agent Name] from [Your Company Name]. I'm calling about your application for the [Job Role] position. Is now a good time for a quick chat?"

For live interaction, the system would use a voice API like Twilio to manage calls. If a candidate accepts, the system could transfer the call directly to an assigned recruiter's phone line. If a callback is requested (e.g., "call me back at 5 PM"), real-time transcription would capture the response, and the system would schedule the callback and update the ATS with relevant notes. All interactions, including full transcripts, would be logged using tools like structlog for auditing and analysis.

The entire system would be delivered as infrastructure as code, deployed in a containerized environment. This approach ensures repeatability and maintainability. Syntora would implement monitoring, such as Slack alerts for API error rates, to ensure the system's operational reliability post-deployment. The client would receive the full codebase, architectural documentation, and guidance on operational handover.