Latest Blog
AI Phone Screens for Staffing Agencies: What to Test Before You Buy
Researcher
•
5 min read
Share this post
Bullhorn surveyed nearly 2,300 recruitment professionals in November and December 2025 and found that 56% of its highest-growth firms place candidates in under 10 days. For comparison, SHRM's 2026 benchmarking data, from 4,657 respondents surveyed between 24 November 2025 and 23 January 2026, puts the median time to fill a nonexecutive role at 39 calendar days. Agencies win work by being faster than that, and most of the difference is decided in the first two days, before anybody has submitted anything.
So the question worth asking about an AI phone screen is narrow. Does it get a usable answer out of a candidate sooner than a recruiter would have, and does the recruiter end up with fewer hours on that desk at the end of the week?
Where do the first two days go?
They go into reaching people. A recruiter working a light industrial desk has a list of applicants who are on a shift, driving, or asleep because they worked nights, and the recruiter is available between nine and five. Every hour of that mismatch is an hour a competing agency can use.
The second thing eating those days is repetition. The same six or seven questions decide whether somebody goes forward on most high-volume roles. Can they get to the site, do they have the certification, will they take the shift pattern, do they have the right to work, what is their pay floor, when can they start. A recruiter asking those on 40 calls a day is doing skilled work about twice.
Bullhorn's survey found that 46% of firms said AI cut screening time by half or better and 55% said AI screening improved their KPIs by more than 25%. Those are self-reported figures from firms that chose to adopt, so read them as what adopters believe they got, not as a measured effect.
Book a Desk Throughput Review. Thirty minutes on where your time to first contact goes on one desk, using your own Bullhorn or ATS data. We will do the arithmetic with you whether or not you buy anything.
What does a phone screen have to produce to be worth automating?
It has to produce a decision a recruiter can act on in seconds. A transcript that a recruiter still has to read end to end has moved the work without removing it, and pilots die in week six on exactly that.
Ask any vendor to show you the output for one of your own roles, and spend the demo on that rather than on the conversation. Does it come back as the six answers your recruiter needs, in fields, in the order the submission requires? Does it write into your ATS as structured data or as a note somebody has to retype? Can a recruiter decide from it in under thirty seconds without opening a second application?
Then ask for throughput before and after, on a customer's own numbers, for a desk that looks like yours. Submittals per recruiter per week, time from application to first contact, and time from first contact to submission. A vendor who can produce those three for a comparable customer is telling you something. A vendor who can only produce a demo is telling you something too.
Does AI adoption pay in staffing?
The evidence points that way and it cannot carry the weight people put on it. Bullhorn found that its top-performing firms were four times more likely to use AI, and that 78% of firms growing revenue by 25% or more used AI tools inside their ATS. Both are cross-sectional correlations from a survey of a software vendor's own market, taken at one point in time. Fast-growing firms buy more of everything, including software, so some of that arrow points the other way.
The adoption picture is thinner than the headline suggests. Only 10% of firms in the same survey had agentic AI implemented across the full workflow. Most of the market is running one or two automated steps inside a process that is otherwise the process it had in 2023.
Cost pressure is the part that is not in dispute. Appcast's tenth annual benchmark report, released on 17 February 2026 and drawn from 302 million clicks and 27 million applications across nearly 1,200 employers, found cost-per-application and cost-per-hire both rose sharply through 2025. If each applicant costs more to acquire, every applicant who never gets a call is a larger write-off than they were last year.
Book a Pilot Scoping Call. Forty-five minutes on which one of your desks would give you a clean read in six weeks, and what you would need to measure to trust the result.
What happens if you submit somebody who is not who they said they were?
Your client finds out, not you, and the whole problem sits in that sentence. An agency carries the exposure for identity in a way a direct employer does not, because the person who discovers the fraud is the customer you want to renew with.
Gartner has predicted that one in four candidate profiles worldwide will be fake by 2028, and in the same work, reported by HR Dive on 8 August 2025, 6% of the 3,000 candidates surveyed admitted to taking part in interview fraud. That reporting is a year old and the underlying behaviour has not become rarer since.
Voice-only screening sits awkwardly here. A phone call gives you no face to check, which is why identity verification has to happen somewhere in the sequence and has to happen before submission, not after placement. Whatever you use, know which step in your own process would catch a substitution between the screen and the first day on site, and make sure somebody owns that step by name.
Book a Submission Integrity Walkthrough. We will map where identity gets checked in your current sequence and where a swap would go unnoticed. Bring one client account you would least like to lose.
How should an agency pilot this?
Run one desk and one role family for six weeks against numbers you already have in your ATS, and write down the stopping rules before you start.
What to measure | Where the baseline comes from | What would make you stop |
|---|---|---|
Hours from application to first contact | ATS timestamps, last quarter, same desk | Median moves less than half a day |
Share of invited candidates completing the screen | Same desk, recruiter-run calls | Completion falls below your recruiters' rate |
Submittals per recruiter per week | Consultant activity report | Submittals flat while review time rises |
Client rejection reasons on submitted candidates | Client feedback log | Rejections shift toward things the screen was meant to catch |
Watch that last row hardest. An agency screen is only good if the client's rejection reasons change, because those reasons are the only measure you have of whether the qualification was correct as well as fast.
Book a Six-Week Pilot Design Session. We will build the four rows above against one of your desks, including the numbers that would tell you to stop. You keep the design either way.
What we cannot tell you
We cannot tell you what completion rate you will get, and neither can anybody else. Every published figure in this category, including ours, comes from one vendor's mix of clients, roles and regions.
Nothing here fixes a requisition your client has not committed to filling. Agencies burn a large amount of screening effort on roles that were never going to be filled by anyone, and a faster screen makes that waste arrive sooner.
An AI phone screen also does nothing for the client relationship. Account managers win and keep accounts. Software does not.
Where Tenzo fits
Tenzo runs the structured first-round screen for staffing firms. It sits alongside Bullhorn or whatever ATS you run, calls or texts candidates as soon as they apply, works through the qualification questions for that role, and hands a recruiter structured answers with the full transcript underneath. Interview design is configurable by role, accommodation paths are documented, and there is version history for audit. Humans review throughout and make all final decisions.
Of candidates who apply to a role and are invited to interview, 80% go on to complete an interview. Our denominator there is candidates we invited, not everyone who applied. Split it on one of your own desks before you plan against it.
Language shows up more in staffing than most buyers expect. Between 10% and 15% of the blue-collar interviews we run happen in a language other than English, with Spanish the most common. Candidates interviewing in their own language rate the experience higher and opt out less often than they do in English. We are not publishing a percentage on that second point, because what we have is consistent and qualitative.
On identity, Tenzo verifies the candidate during the AI interview, looks for signs of cheating, and looks for signs that the candidate has been swapped for somebody else. On the application itself it checks the phone number, the IP address, the physical address, the email history and the other records available for signals of fraud. None of that beats meeting somebody in person, ours included. Where a client account is worth enough to justify the cost, put a person in front of the candidate before the first shift.
To find out whether an AI screen would change your time to submit, book a working session. Bring one desk and last quarter's submittal numbers.
FAQ
How fast do the best staffing firms place candidates? Bullhorn's 2026 GRID report, from nearly 2,300 recruitment professionals surveyed in November and December 2025, found 56% of its highest-growth firms placing in under 10 days. SHRM's 2026 benchmarking puts the median corporate time to fill a nonexecutive role at 39 calendar days, which is the speed agencies are usually selling against.
Does AI adoption cause faster growth in staffing? Nobody has shown that. Bullhorn found top-performing firms four times more likely to use AI and 78% of firms growing 25% or more using AI inside their ATS. Those are correlations from one survey at one point in time, and fast-growing firms buy more software for many reasons.
Phone or video for a first screen? For high-volume hourly, shift-based and field roles, phone reaches more people, because answering a call needs no app, no camera and no quiet room. Video gives you a face to check against the identity document. Plenty of agencies run phone first and put the face check later in the sequence, before submission.
What should we measure in a six-week pilot? Hours from application to first contact, completion among invited candidates, submittals per recruiter per week, and the reasons clients give for rejecting the candidates you submit. Take all four from your ATS for the same desk last quarter and agree the stopping rules up front.
How big is the candidate fraud problem for agencies? Gartner has predicted one in four candidate profiles will be fake by 2028, and 6% of the 3,000 candidates it surveyed admitted to interview fraud, reported in August 2025. For an agency the stake is higher than for a direct employer, because your client is the one who finds out.
