Automate Recruiting Reference Checks with a Custom Voice AI
Yes, a small business should use a voice AI consultant to build a custom reference checking system. This automates calls, transcribes responses, and flags key insights, saving recruiters hours per candidate.
The system's scope depends on the number of questions, required integrations with your Applicant Tracking System (ATS), and language support. A five-question script in English that connects to Greenhouse is a standard build. A multi-language system with question branching based on candidate seniority requires more development.
We built a reference checking system for a 12-person recruiting firm processing 400 applicants per month. Their team spent 30 minutes per candidate on manual calls and scheduling. The new system conducts and analyzes three reference calls in under five minutes total, and went live in four weeks, cutting their average time-to-hire by two days.
What Problem Does This Solve?
Recruiting teams start with manual phone calls, which are slow and inconsistent. Different recruiters ask questions differently, and hand-written notes are impossible to search or compare across candidates. This process creates a bottleneck, delaying offers and causing candidates to accept competing ones.
Off-the-shelf software like Checkster or Outmatch tries to solve this but relies on sending web forms to references via email. This approach suffers from low response rates as references ignore the emails. The written answers they do collect lack the nuance and candor of a spoken conversation. These platforms also charge per-hire or per-seat fees that become expensive for a growing firm.
Technically-inclined teams might try a visual IVR builder like Twilio Studio. While it can place calls and play audio prompts, it cannot handle the critical task of understanding unstructured verbal responses. A simple "press 1" workflow is easy, but analyzing a two-minute answer about a candidate's weakness requires a custom speech-to-text and language model pipeline, which cannot be built within a drag-and-drop interface.
How Does It Work?
We begin by mapping your exact reference questions into a structured script and provision a dedicated phone number using the Twilio API. We then connect directly to your ATS, whether it is Greenhouse, Lever, or Ashby, using its API and a Python script with httpx to pull candidate and reference contact details automatically. This setup eliminates all manual data entry for your recruiters.
The core system is a Python application built with the FastAPI framework and deployed on AWS Lambda for event-driven execution. When a candidate reaches the reference check stage in your ATS, a webhook triggers the Lambda function. It places outbound calls, plays pre-recorded audio prompts for each question, and records the reference's spoken responses, allowing up to 90 seconds of audio per question.
Each audio recording is sent to Anthropic's Claude 3 Sonnet API. We use a specific prompt chain that instructs the model to perform three tasks: generate a verbatim transcript, write a concise summary of the entire call, and check the transcript for predefined keywords like "conflict" or "leadership". This entire analysis, from call completion to structured data output, finishes in under 15 seconds.
The final package containing the summary, transcript, and a link to the original audio recording is posted as a private note on the candidate's profile in your ATS. This is written back using the ATS API. The system is built to handle up to 10 concurrent calls, and monthly hosting costs on AWS Lambda and Supabase are typically under $50 for processing 500 reference calls.
What Are the Key Benefits?
From a 30-Minute Call to a 3-Minute Summary
Automated calls run 24/7 without recruiter involvement. Get structured notes and audio links back in your ATS in under three minutes after the reference hangs up.
One Fixed-Price Build, No Per-Seat Fees
You pay for the initial scoped build, not a recurring subscription that scales with your team size. Your only ongoing cost is for the underlying cloud services you use.
You Receive the Full Python Source Code
The complete codebase is delivered to your company's GitHub repository. You can modify scripts or analysis prompts without vendor dependency or change request fees.
Get Failure Alerts in Seconds, Not Days
We implement structured logging with structlog. If an API call fails or a transcript is garbled, an alert with the specific error is sent to a Slack channel immediately.
Results Appear Directly Inside Your ATS
We push summaries and audio links to candidate profiles in Greenhouse, Lever, or Ashby. Your team gets the data they need without learning a new tool or dashboard.
What Does the Process Look Like?
Discovery and Integration (Week 1)
You provide your reference check script and grant read/write API access to your ATS. We map the workflow and confirm the data fields for integration.
Core Voice Agent Build (Week 2)
We build the calling, recording, and analysis logic with Python, FastAPI, and the Claude API. You receive a demo to test the call flow with your own phone.
Deployment and Testing (Week 3)
We deploy the system to your AWS infrastructure and connect it to your live ATS. Your team runs 10-20 real reference checks to validate the end-to-end process.
Handoff and Monitoring (Week 4+)
After a final round of prompt tuning, you receive the full source code and a runbook. We actively monitor system performance for 30 days post-launch to ensure stability.
Frequently Asked Questions
- How is the cost determined for a custom reference check system?
- Pricing is based on three factors: the number of ATS integrations, the complexity of the call script (e.g., conditional logic), and the number of languages required. A standard English-only system for one ATS is a 3-4 week build. Multi-language support or complex branching can extend the timeline. We provide a fixed-price quote after our discovery call.
- What happens if a reference's voicemail picks up?
- The system has voicemail detection that analyzes the first 3-5 seconds of audio. If it detects a common voicemail greeting, it hangs up and retries the call 2 hours later, up to three times. After the third failed attempt, it flags the reference in your ATS for manual follow-up by a recruiter, ensuring no candidate gets stuck.
- How is this different from using a service like Callingly?
- Callingly is a live-call routing tool for sales teams to connect with leads instantly. Our system is an asynchronous data collection agent for recruiting. It doesn't connect two people. It autonomously calls references, records their unstructured answers to specific questions, and uses a language model to analyze the audio for your team.
- How do you handle the privacy of reference data and recordings?
- The entire system is deployed on your own cloud infrastructure, typically AWS. Syntora has no access to recordings or transcripts after the project handoff. Audio files are configured with a 30-day auto-delete policy by default to comply with data retention rules. All data processing is contained within your environment, giving you full control and ownership.
- How accurate is the transcription and analysis?
- The Claude API provides transcription with over 95% accuracy for clear English audio. The quality of the analysis depends on the clarity of your questions. During the testing phase, we work with you to refine the analysis prompts, ensuring they reliably extract the specific insights your recruiters need from the conversations.
- Can we change the questions or the voice after the system is built?
- Yes. The questions are stored in a simple configuration file that you can edit. The system uses a text-to-speech service to generate the audio prompts automatically, so any text changes are reflected in the next call. You can also select from dozens of voices and accents. The runbook we provide includes instructions for making these changes without modifying code.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call