AI Phone Screens: How Agencies Screen 100 Candidates a Day

Screening 100 candidates a day means roughly 2,200 screens a month on a 22-working-day month. At the pace of about twenty calls per recruiter per day that Trembit reports for an unnamed staffing client, that is five recruiters on calls all day. AI voice agents are being sold as the way to cover that volume without hiring five people, and a 70,000-applicant field experiment found AI-led interviews produced 12% more job offers in one high-volume setting.

This post covers what the evidence shows, what the volume costs at published prices, and which laws to check first. The volume figures below are arithmetic, not a case study of a named agency.

Can AI really screen 100 candidates a day?

Technically yes: vendors sell agents built for far more than that. Alex's homepage lists a phone channel and says it will let Alex call thousands of candidates instantly, and ConverzAI's homepage claims 3x more placements, 90% faster submissions, 99% engaged candidates and 85% less manual work for staffing firms. Those are company claims with no published methodology.

The harder question is whether the output is good enough to act on. Trembit's case study, also vendor-written, says the client's screens were 15-minute calls where each recruiter handled about twenty a day, and that the bot cut the backlog to same-day turnaround. Trembit builds custom voice bots, so it has a commercial interest in that result.

What does the best evidence say about AI voice interviews?

The best public evidence is a randomised field experiment by Brian Jabarian and Luca Henkel, posted on arXiv (version 2 dated 11 September 2026). Applicants were randomly assigned to human recruiters or AI voice agents, and the AI-interviewed group was 12% more likely to receive job offers, with higher job starts and retention and no decline in the productivity of hired workers.

Key details from the full paper:

• Setting: 70,884 applications for entry-level customer service jobs at PSG Global Solutions, a recruitment process outsourcing firm and Teleperformance subsidiary.
• Design: applicants who passed an initial screening were randomly assigned to a human, to an AI voice agent, or allowed to choose. Human recruiters made all hiring decisions.
• Outcomes: the AI group had an 18% higher likelihood of starting the job.
• Candidate preference: when given the choice, 78% picked the AI voice agent.
• Caveat in the paper: AI-led interviews were rated significantly less natural, and applicants who chose the AI had lower language and analytical scores than those who chose a human.

The authors attribute the gains to more structured and consistent interviews, not to the AI judging better. That matters for agencies: the benefit comes from every candidate getting the same questions.

What does 100 screens a day cost at published prices?

At Ribbon's published prices, 2,200 interviews a month costs about $4,999 on the Scale plan, or roughly $2.27 per interview. Ribbon is one of few vendors that shows its prices, so it is the only one where the arithmetic can be done from a public page. Plan prices below are the monthly figures, billed annually, as of 16 October 2026 per Ribbon's pricing page.

The cost column is my own arithmetic: plan price plus overage on interviews above the allowance, assuming 22 working days. For Scale, that is $1,999 plus 1,200 extra interviews at $2.50 ($3,000). Ribbon's list shows the interviewer as video and voice in 10+ languages, so check that phone-only calls are covered by your plan.

The active-role cap is the catch for agencies. Scale allows 10 active roles, and a desk running many open requisitions may exceed that, which pushes you toward the custom Enterprise plan.

Which vendors publish prices, and what do others charge?

Few do. Culturro's September 2026 review of 16 AI interview agents found that three publish a price: Ribbon, CodeSignal and Employment Hero. The rest quote from headcount or hiring volume.

For Alex (formerly Apriora), Culturro reports that annual agreements start at $36,000, with the final price depending on interview volume, connected systems and roll-out complexity. That figure comes from Culturro only. Alex says it supports 33+ officially supported ATS integrations.

If you are still choosing between interview platforms, our comparison of Paradox and HireVue covers the enterprise end, and the video interview tools post covers time-to-hire claims.

Which laws apply to AI phone screens?

Four rules are worth checking before you launch: NYC Local Law 144, Colorado's SB26-189, the EU AI Act and the US phone-consent rules. This is general information, not legal advice.

Three details change how you apply these.

NYC. The law defines a covered tool as one that issues a score, classification or recommendation used to substantially assist or replace discretionary decision making. A voice agent that only gathers answers for a recruiter may sit differently from one that ranks candidates. Foley Hoag reports fines of $500 to $1,500 per day of noncompliance.

Colorado. Constangy says the statute excludes tools used solely to summarize, organize, translate, draft, route or present information for human review. It also says a tool that scores or ranks applicants is likely covered, and the law has no private right of action; the Attorney General enforces it.

Phone consent. Wiley's summary says callers must obtain prior express consent or prior express written consent, identify themselves and offer opt-out rights before AI-generated voice calls. How that applies to a candidate who applied online depends on the call, so get counsel's view.

For a lighter overview of the tools around this workflow, see the ATS features staffing agencies use and the headhunter toolkit.

How should an agency set up AI phone screens?

Start narrow, keep a person in the decision, and log everything. These steps follow the structure of the PSG study, where AI collected information and humans decided.

1. Pick one high-volume role. The evidence base is entry-level customer service. Do not assume it transfers to executive or technical search.
2. Write the screening questions first. The study credits structure and consistency, so use the same questions and scoring rubric for every candidate.
3. Time your current process. Measure minutes per screen and screens per recruiter per day before you buy. Trembit's twenty a day is one vendor's figure, not a benchmark.
4. Confirm consent and notice. Decide how candidates agree to an AI call and where you tell them an AI is involved, then check this against the rules above.
5. Keep humans deciding. Culturro found five of the 16 vendors can move or reject a candidate with no recruiter looking first. Check which setting is on before you sign.
6. Route results into your ATS. Scores and transcripts should land where recruiters already work, not in a separate inbox.
7. Review a sample weekly. Compare AI scores with recruiter judgement on a sample of calls, and keep records. Colorado's law requires three years of them for covered tools.

What's still unclear?

Several things the public record does not settle.

• Does it transfer to staffing? The field experiment was at an RPO firm for entry-level customer service jobs. It does not test contract staffing, healthcare or executive search.
• Possible conflict of interest. The paper discloses that data collection finished on 7 June 2025 and that, on 1 August 2025, Jabarian accepted an unpaid role as Chief Economist at PSG Global Solutions. It also states the partner firm had no role in the analyses or the decision to publish. The paper is a working paper posted on arXiv.
• Candidate quality. AI-led interviews were rated less natural, and applicants who picked the AI had lower language and analytical scores. Whether that affects your client's candidates is unknown.
• Vendor numbers. Trembit, ConverzAI and Alex figures are company claims, not independently verified results.
• Pricing gaps. Most vendors do not publish prices. The $36,000 Alex figure comes from one reviewer.
• Signing date conflict. One law firm, Lathrop GPM, gives 20 May 2026 for Colorado's signing, while the governor's office says 14 May 2026. This post uses the governor's date.
• Audit duty. Culturro's review says no vendor carries your audit duty. If you are covered by NYC's law, the bias audit obligation stays with you.