Automate Your 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 replaces hours of manual recruiter calls with a consistent, 5-minute automated process per candidate.
The scope depends on your existing systems. Connecting to an Applicant Tracking System (ATS) like Greenhouse via its API is straightforward. Adding custom sentiment analysis or keyword flagging requires access to your historical reference check notes to fine-tune the AI's prompts.
We built a system for a 12-person recruiting firm processing 400 applicants per month. Their three recruiters spent over 30 minutes per candidate on manual reference calls. The system launched in two weeks, reducing active recruiter time to just three minutes of review per candidate.
What Problem Does This Solve?
Recruiting teams often try using generic calling services. A tool like Twilio Studio can place calls, but it cannot interpret the unstructured audio responses or summarize the results into a candidate profile. You get a raw MP3 file, which a recruiter still has to listen to, defeating the purpose of automation.
A common scenario involves a 15-person firm trying to scale from 20 to 50 hires a year. Their two recruiters are a bottleneck. Reference checks for 5 finalists take 10-15 hours of phone tag and note-taking. They look at enterprise recruiting platforms like HireVue, but the $20,000 annual contract and per-seat licensing model are designed for 500-person companies, not them.
These off-the-shelf tools fail because the core problem is not placing the call; it is understanding the answer. They force a choice between a raw audio file that creates more manual work or an expensive, inflexible AI module that cannot be customized to ask role-specific questions. They provide a partial solution that does not address the actual bottleneck: recruiter time spent on low-value, repetitive conversations.
How Does It Work?
We begin by mapping your existing reference questions into a structured script. We use the Claude API to generate dynamic, natural-sounding follow-up questions based on the reference's initial answers. The system integrates with your ATS (e.g., Lever, Greenhouse) using Python and the httpx library to pull candidate and reference contact information automatically.
The core calling logic is built on AWS Lambda using Twilio's API for programmable voice. When a candidate reaches the reference check stage in your ATS, a webhook triggers the Lambda function. The function initiates an outbound call, asks the scripted questions, and records the reference's responses. The audio file, typically 2-4 minutes long, is saved directly to a secure Amazon S3 bucket in your account.
A second AWS Lambda function triggers the moment the audio file is saved. It uses an audio-to-text model to produce a full transcript, a process that takes about 15 seconds for a 3-minute call. That transcript is then sent to the Claude API with a detailed prompt engineering sequence. This prompt instructs the AI to write a concise summary, score responses against key competencies, and extract any predefined keywords.
The entire analysis pipeline executes in under 60 seconds. The final summary, competency scores, and full transcript are posted back to the candidate's record in your ATS via an API call. We use structlog for detailed logging, and a failed call or low-confidence transcription (below 90%) sends an immediate alert to a designated Slack channel. The system can handle up to 50 concurrent calls, and monthly AWS hosting costs are typically under $20.
What Are the Key Benefits?
Get Summaries in 60 Seconds, Not Hours
The entire process from call completion to a summarized report in your ATS takes less than one minute. Recruiters review insights, not listen to recordings.
Pay For The Build, Not Per Recruiter
A one-time project cost with minimal monthly hosting. This avoids the high per-seat fees of enterprise recruiting platforms that charge based on headcount.
You Own the Code and the Prompts
You receive the full Python source code and the exact Claude API prompts we developed. There is no vendor lock-in; your asset is portable and resides in your GitHub.
Failure Alerts Sent Directly to Slack
We configure webhooks to send real-time alerts if a call fails or transcription quality is low. You know instantly if something needs manual attention.
Writes Directly to Greenhouse or Lever
The system uses your ATS's API to read candidate data and write back summaries. It becomes a native part of your existing recruiting workflow, no new software to learn.
What Does the Process Look Like?
Script and ATS Access (Week 1)
You provide your standard reference questions and create an API key for your ATS. We draft the initial call script and map the necessary data fields for integration.
Core System Build (Week 2)
We build the calling and transcription pipeline on AWS Lambda. You receive access to a staging environment to test calls with your team's phone numbers.
Integration and Testing (Week 3)
We connect the pipeline to your live ATS and test the full workflow. You receive a draft of the system runbook for review and feedback.
Launch and Monitoring (Week 4+)
The system goes live. We monitor performance for 30 days to handle edge cases. You receive the final source code, documentation, and full ownership.
Frequently Asked Questions
- How is the project cost determined?
- Cost depends on three factors: the number of custom question scripts, the complexity of the ATS integration, and the level of custom analysis required. A basic system that transcribes and summarizes can be built in two weeks. Adding sentiment analysis and keyword extraction for 10-15 keywords might extend the timeline to three weeks.
- What happens if a reference's voicemail picks up?
- The system detects voicemail tones with over 95% accuracy. When detected, it hangs up and schedules a retry call for two hours later. After three failed attempts over 24 hours, it flags the reference in the ATS for manual follow-up by a recruiter, ensuring no candidate gets stuck in the process.
- How is this better than using a virtual assistant (VA) service?
- A VA performs the task manually, introducing human inconsistency and scheduling delays. They bill by the hour, which scales poorly with hiring volume. Our system runs 24/7, provides a perfectly consistent experience for every reference, and costs a fraction of a VA's hourly rate to operate once built. You are buying an asset, not renting time.
- How is the reference's data and privacy handled?
- All data is processed within your own AWS account. Audio files and transcripts are stored in your S3 bucket, and you control the data retention policies. Syntora does not have ongoing access to your production data after the initial 30-day monitoring period. The system is designed to be fully self-contained within your infrastructure.
- Can the AI ask different questions for different roles?
- Yes. The system can pull the job title or department from your ATS. We build a simple mapping that directs the AI to use a specific question script. For example, a 'Sales' role gets questions about quota attainment, while an 'Engineering' role gets questions about technical collaboration. This is managed in a simple configuration file.
- Does it sound like a robot?
- We use modern text-to-speech APIs that are nearly indistinguishable from human speech, with natural inflection and pacing. You can choose from dozens of voices and even adjust the speaking rate. The goal is a professional and comfortable experience for the reference, not a jarring, robotic interaction. We test this with you during the build to match your company's tone.
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