Screening hundreds of candidates is a massive time sink. Most hiring managers spend hours on the phone asking the same five questions, only to realize within the first minute that the candidate isn't a fit.

We recently built a system that handled 895 calls and 2,800 minutes of phone screening automatically. The average cost? 30 cents a call.

In this guide, you can learn exactly how to build a Voice AI agent that calls candidates immediately after they apply, asks qualifying questions, and logs the results in a spreadsheet. You only ever have to speak to the candidates who are actually qualified.

Check Video Guide Here:

‍

The Core Concept: How It Works

The workflow is simple but powerful. It replaces the "first round" phone screen with an autonomous agent.

1. Application: A candidate submits a form (name and phone number).
2. The Call: The system automatically calls them.
3. The Interview: An AI agent (we call her "Sarah") asks 5 qualifying questions.
4. The Log: The answers are transcribed, analyzed, and sent to a Google Sheet.
5. The Verdict: The sheet automatically marks them as "Qualified" or "Not Qualified."

The Tech Stack

• Retell AI: Handles the voice infrastructure (Speech-to-Text, LLM processing, Text-to-Speech).
• n8n: The automation platform that connects the AI to your database.
• Google Sheets: Acts as the database and dashboard.

‍

Understanding the "Voice Loop"

Before building, it is helpful to understand what is happening under the hood. AI voice isn't magic; it is a rapid sequence of events that happens in roughly 1 second.

1. Caller Speaks: The audio is captured.
2. Speech-to-Text (STT): The audio is converted to text.
3. The Brain (LLM): The AI thinks of a response based on your prompt.
4. Text-to-Speech (TTS): The text is converted back to audio.
5. Caller Hears: The candidate hears the response.

‍

We use Retell AI because it orchestrates this entire loop for us with ultra-low latency, costing roughly $0.10/minute.

Step 1: Configuring the AI Agent (Retell AI)

First, we need to create the brain of the operation.

1. Create an Agent: In Retell AI, create a new "Single Prompt Agent."
2. Select a Voice: Choose a voice that fits your brand. We used "Hailey" (American, young) for a friendly vibe.
3. The Prompt: This is the most critical part. You must define the persona and the objective.

Example Prompt Snippet:

"You are Sarah, a recruiting assistant for Metro Fulfillment Services. Your role is to ask five screening questions. Be friendly, warm, and efficient. First, ask if they have time for a 2-minute screening..."

Data Extraction

You don't just want a transcript; you want structured data. In the Post Call Data Extraction settings, define the specific variables you want the AI to pull from the conversation.

We set up the following "Boolean" (True/False) variables:

• lift_50_pounds
• reliable_transportation
• overnight_weekend_availability
• authorized_to_work

This ensures that even if the candidate says, "Yeah, I guess I can do that," the AI converts it to a clean TRUE or FALSE for your database.

Step 2: Building the Automation (n8n)

Now we need to get the data out of Retell and into a spreadsheet. We use n8n for this because of its flexibility with Webhooks.

1. The Webhook: create a "Webhook" node in n8n. This gives you a URL to paste into Retell AI.
2. The Trigger Event: This is a common pitfall. Retell sends signals when a call starts, ends, and is analyzed. You must configure n8n to listen for the call_analyzed event. This is the only event that contains the extracted data (the answers to your questions).
3. Test the Connection: Run a test call. You will see a JSON packet arrive in n8n containing the phone number, recording URL, and the extracted variables (e.g., lift_50_pounds: true).

Step 3: The Dashboard (Google Sheets)

Create a Google Sheet with columns matching your data points: Name, Phone, Lift 50lbs, Transport, Overnight Shift, Qualified?

1. Connect Sheets to n8n: Add a Google Sheets node in n8n.
2. Map the Fields: Drag and drop the data from the Webhook node into the Google Sheets columns.
3. The Qualification Logic: In Google Sheets, you can use a simple formula to determine if a candidate is qualified.

Step 4: Go Live

You have two ways to trigger these calls:

1. Batch Mode: Upload a CSV of 100 applicants to Retell AI and click "Send." The system will call them all simultaneously (or scheduled).
2. Real-Time: Connect your application form (like Typeform) to the system so the call happens the second they hit "Submit."

Instead of spending your day leaving voicemails, you open your spreadsheet. You see 50 rows. 40 are marked "False" (unqualified). You only call the 10 people marked "True."

Migration is a business decision. You aren't replacing the human connection; you are optimizing it so you only connect with the people who matter.

‍

If you want to implement this system but don't have the time to configure the Webhooks and LLM prompts yourself, we can help. Contact us!

‍