How to Set Up AI Candidate Screening Calls: A Step-by-Step Playbook for Recruitment Agencies

What this playbook covers

The recruitment overview explains what voice AI does for an agency; the candidate screening page explains the product. This is the how-to: what to prepare before the first automated call, how to write screening questions a phone conversation can actually carry, how to score answers without pretending the AI is a recruiter, and how to hand a qualified candidate to a human with the context intact.

The Screening Bottleneck in Recruitment

A typical recruitment agency receives 100-500 applications per open position. Recruiters must call each candidate to verify basic eligibility — experience, current salary, notice period, location preference.

This manual screening takes 5-10 minutes per candidate. For a team handling 20 positions, that's 2,000-10,000 screening calls per month. Most agencies simply can't keep up, leading to:

• Candidates accepting other offers while waiting for your call
• Inconsistent screening quality across recruiters
• Best candidates slipping through the cracks
• Recruiters burning out on repetitive calls

How AI Screening Works

FrontivaAI's AI candidate screening automates the entire initial screening process:

1. Upload candidate list — CSV with names and phone numbers
2. AI calls each candidate — natural conversation in Hindi or English
3. Asks screening questions — experience, skills, salary, notice period, location
4. Scores and ranks candidates — based on your criteria
5. Schedules interviews — for qualified candidates directly with recruiters
6. Sends summary report — with all responses, scores, and recordings

All this happens in parallel — the AI can screen 100 candidates in 2 hours vs 2 weeks manually.

Results: Faster Time-to-Fill

→ See the full recruitment AI use case

Step 1: decide what a screen is allowed to decide

Before a single call is placed, write down the floor for the role - the facts a candidate must meet to be worth a recruiter's time - and nothing else. Years of experience in the function, current and expected CTC, notice period, city and willingness to relocate, shift availability, and one or two role-specific facts (a language, a licence, a tool). That list is the screen. Everything that is a judgement - communication, attitude, fit - is not, and putting it in the script only makes the AI pretend to assess something it cannot.

Set the thresholds in the dashboard, not in the questions. The agent asks every candidate the same way; the filter is applied to the recorded answers afterwards, so a recruiter can loosen a threshold on Friday without re-calling anyone.

Step 2: write questions a phone call can carry

Questions written for a form fail on a call. Three rules fix most of them:

• One fact per question. "How many years have you worked in inside sales?" - not "tell me about your sales background".
• Ranges where a number is awkward. Expected CTC is asked as "roughly, in lakhs per annum"; the agent records the number the candidate gives without haggling.
• Role-specific last. A candidate who fails the floor should not be asked to explain a technical process for five minutes first.
• Always an ending. Either the interview offer with two slots, or "the recruiter will call you" - never a dead line.

In Hindi and English the questions are the same; the agent follows the candidate's language from the first full sentence, which matters for field-sales and BPO roles where the honest answer comes in Hinglish.

Step 3: the hand-off to a recruiter

The output of a screen is a list, sorted by the answers, with every call's transcript one click away. The recruiter reads the top of the list, not the whole thing, and books the interviews the agent has not already booked. Three things make this work in practice:

• Structured fields, not notes. "Notice period: 30 days" is sortable; "he said about a month I think" is not.
• The interview goes into the calendar the recruiters share, with a WhatsApp confirmation to the candidate and a reminder before the slot - interview no-shows are the afternoon-eater of every agency.
• Escalation is explicit. A candidate who asks something the script does not cover - "is the role remote?" when you have not said - gets "the recruiter will confirm that", and the transcript flags it.

Week-one mistakes, and how to avoid them

Agencies that struggle in the first week usually made one of four choices:

• Calling at the wrong hour. Working candidates answer after 7 pm and at lunch; the agent can call at any hour, so schedule it for the hours candidates pick up rather than the hours the office is open.
• Screening for the wrong floor. A threshold set too tight produces an empty list and the conclusion that "AI does not work"; start loose, tighten after you have seen fifty transcripts.
• No retry rule. A first call that goes unanswered is not a rejection; set two retries at different hours and a WhatsApp in between.
• Hiding that it is an AI. The agent introduces itself as the agency's AI assistant on every call. Candidates who know they are talking to software give plain answers; candidates who feel tricked hang up.

Outbound screening calls in India also carry TRAI obligations - identification at the start, respect for the do-not-disturb registry, no calling numbers you have no relationship with. Applicants have given you their number; strangers from a scraped list have not.

Sources and further reading

Everything above about FrontivaAI itself is stated from our own product and price list. Where we lean on an outside fact, this is where it comes from - and if a figure matters to your decision, check it at the source rather than trusting our summary.

• TRAI - Telecom Commercial Communications Customer Preference Regulations, 2018 (the rules for commercial calls and the DND registry)
• WhatsApp Business Platform documentation - message templates, which is how a business can message a customer who has not written first
• Digital Personal Data Protection Act, 2023 - Ministry of Electronics and IT