How AI Voice Agents Cut Time-to-Submit for US Staffing Agencies by 60%
Key takeaways
- Time-to-submit is the clock from job order received to first qualified candidate submitted. It is the metric your client’s VMS grades you on, and the one almost nobody on your team is timing.
- No independent research body publishes a time-to-submit benchmark. SHRM benchmarks time-to-fill, at a median of 39 calendar days for nonexecutive roles across more than 4,600 organisations. Nobody benchmarks the agency-side number.
- Break a typical contract submit into five segments and roughly 73% of the elapsed hours sit in one of them: the outreach and screening queue. That is queue time, not work time.
- A voice agent removes the queue by calling concurrently instead of serially. In our model that takes 33 working hours down to about 12, a 63% reduction. We round it down to 60% in the headline.
- The 60% is a ceiling, not a promise. It only lands if your queue really is three quarters of your clock. Baseline your own five timestamps for four weeks before you buy anything.
The number your client is grading you on
Somewhere in your client’s vendor management system there is a scorecard with your name on it.
It lists how many requisitions you were released, how many you responded to, how fast you responded, and how often your submissions turned into interviews. Your account manager may have seen it once. Most agency owners never have.
And the column that moves your position on that scorecard more than any other is not fill rate. It is how many hours passed between the requisition hitting your inbox and your first qualified candidate landing in the portal.
Forty vendors got that req. The scorecard is a race, and it is being timed whether or not you own a stopwatch.
This article is about that stopwatch. Where the hours actually go, which of them automation can remove, and how to work out your own number instead of borrowing anybody else’s. Including ours.
This is a companion piece to our guide to AI voice agents for staffing agencies, which covers the whole category. This one takes a single metric apart.
What time-to-submit actually measures
Time-to-submit is the elapsed time between receiving a job order and submitting your first qualified candidate to the client. The clock starts when the requisition arrives, not when you decide to work it. It stops on submission, not on client acknowledgement. It is the only recruiting metric an agency fully controls.
Two rules make it useful rather than decorative.
Start the clock on receipt. Not on qualification, not on the moment a recruiter picks it up. Your client’s clock started when they released the req, so yours has to as well. Starting late is how agencies convince themselves they are fast.
Measure in working hours, not calendar days. Days round everything to the nearest disaster. “Two days” hides whether you lost twenty hours to a screening queue or four hours to a portal you could not log into. Hours make the problem visible. Days make it invisible.
Time-to-submit vs time-to-fill vs time-to-hire
These get used interchangeably and they measure completely different things.
Time-to-fill runs from requisition open to offer accepted. Time-to-hire runs from candidate application to offer accepted. Both belong to the employer, and both include weeks of client-side interviewing you do not control.
Time-to-submit is yours. It ends the moment you hand the candidate over, which makes it the only one of the three you can be held to and the only one you can actually fix.
Why your client is already measuring it even if you are not
Every VMS and every MSP programme records a submission timestamp. That is not a feature, it is the entire point of running a vendor panel: rank suppliers on responsiveness and quality, release the next req to whoever ranks well.
So the metric exists in your client’s system whether or not it exists in yours. Not tracking it does not mean you are not being scored on it. It means you are being scored on it blind.
Nobody publishes a benchmark for this, and that is the problem
Here is what makes this article different from every other page on this subject. We went looking for an independent benchmark for time-to-submit, and there isn’t one.
SHRM runs the most comprehensive recruiting benchmarking programme in the US. Its 2026 Recruiting Executives Benchmarking report, published in June 2026 and drawing on more than 4,600 organisations, puts the median time-to-fill for nonexecutive positions at 39 calendar days. Same report: median requisitions per recruiter at extra-large organisations rose 67% in a single year, from 60 to 100. Real numbers, real sample, real methodology.
Josh Bersin with AMS built a Time-to-Hire Benchmark Factbook from half a million data points across eight industries and more than 25 countries.
Neither of them publishes time-to-submit. Neither ever will, because both are built for corporate talent acquisition teams, and a corporate TA team has nobody to submit to. There is no client on the other side of the transaction.
Which leaves the agency metric that decides who wins the requisition with no independent benchmark at all. And a vacuum like that gets filled by vendors, which is why the pages ranking for this term today carry numbers like “80% reduction in time-to-hire” with no source attached to any of them.
So we are not going to hand you a measured average. There isn’t one to hand you. What follows is a model, its inputs are printed on the page, and you should argue with them.
Where the hours go: a time-to-submit decomposition
Take a mid-market IT staffing desk and a contract requisition. Break the work into five segments and put a range of working hours against each.
| Segment | What happens | Working hours | Bounded by |
|---|---|---|---|
| 1. Intake and qualification | Read the req, confirm rate, location, duration, clearance | 1 to 3 | Account manager and client availability |
| 2. Sourcing and matching | ATS search, job board search, building the call list | 2 to 4 | Database quality and search skill |
| 3. Outreach and screening queue | Dialling, voicemails, retries, the screening conversations themselves | 18 to 30 | Recruiter hours |
| 4. Write-up and submittal prep | Formatting the resume, writing the summary, confirming rate | 1 to 2 | Recruiter hours |
| 5. Internal review and submission | Account manager review, VMS or portal entry | 1 to 4 | Reviewer availability |
| Total | 23 to 43, midpoint 33 | A little over four working days |
Why the screening queue is 73% of it
This is the part worth slowing down for, because the number looks wrong at first glance.
Twelve candidates at fifteen minutes each is three hours of talking. So where do eighteen to thirty hours come from?
They come from the fact that segment 3 is not work. It’s a queue.
Three things stack on top of each other. First-attempt connect rates on candidate mobile numbers are low, so most dials produce a voicemail rather than a conversation. Retries have to be spaced out, across hours and across the working day, because calling somebody four times in ten minutes is how you get blocked. And a recruiter holds exactly one conversation at a time while also working three other live reqs, a client call and an interview debrief.
Serial dialling against a low connect rate is the constraint. Not the conversation. What that queue actually costs you in coverage, and which calls are worth handing over first, is the subject of AI candidate screening phone calls.
That is why the segment is 73% of your clock, and it is why every attempt to fix time-to-submit by asking recruiters to work harder fails. You cannot out-hustle a queue. You can only stop it being serial.
What changes when the calls run concurrently
Now run the same requisition with a voice agent handling first contact and screening, and go segment by segment. Being specific here matters, because the honest answer is that most of the segments do not move at all.
| Segment | What the agent changes | Working hours |
|---|---|---|
| 1. Intake and qualification | Nothing. Human work. | 1 to 3, unchanged |
| 2. Sourcing and matching | Nothing. Matching is an ATS problem, not a voice problem. | 2 to 4, unchanged |
| 3. Outreach and screening queue | Calls run concurrently. Retries are automatic and spread into the evening. Throughput stops being bounded by recruiter hours. | 2 to 6 |
| 4. Write-up and submittal prep | Partly. The structured summary comes out of the transcript. A human still checks and formats it. | 0.5 to 1 |
| 5. Internal review and submission | Nothing. Human and portal work. | 1 to 4, unchanged |
| Total | 6.5 to 18, midpoint about 12 |
Segment 3: from serial dialling to concurrent calling
The agent calls two hundred matched records in the first hour instead of one every twelve minutes. Voicemails get retried on a schedule nobody has to remember. Candidates who pick up at 8pm get screened at 8pm, because the agent does not go home.
The ceiling stops being your recruiters’ hours and becomes something you cannot automate away: when candidates are willing to answer their phones. That’s why the segment lands at two to six hours rather than zero.
Segment 4: the write-up comes out of the transcript
Every screening call produces a transcript, a structured set of answers mapped to your fields, and a summary. Your recruiter edits and formats rather than reconstructing the conversation from memory and half a page of notes.
Small saving per candidate. Meaningful across a week.
The three segments a voice agent does not touch
Intake, sourcing and internal review are all untouched in this model, and that is not a gap in the technology. It is what the technology is.
Intake needs a conversation with an account manager. Sourcing is a search and matching problem. Internal review needs a human who is willing to put their name against a submittal.
Anyone telling you a voice agent compresses all five segments is selling. Ask them which segment, and watch what happens.
The 60% is a ceiling, not a promise
Thirty-three working hours down to about twelve is a 63% reduction. We round it down to about 60% in the headline, because the model is built from ranges and precision you have not earned is just another kind of lie.
But the more useful thing is what the number depends on. The reduction is a direct function of how much of your elapsed time sits in the queue. In our baseline that share is 73%. Move it, and the headline moves with it.
| Share of your time-to-submit in the screening queue | Reduction to expect |
|---|---|
| 80% | Around 69% |
| 73% (our model) | Around 63% |
| 60% | Around 52% |
| 45% | Around 40% |
| 30% | Around 27% |
So the honest version of this article’s headline is not “voice agents cut time-to-submit by 60%”. It is this: about 60% of a typical contract submit is queue time, and queue time is the only thing a voice agent removes.
If your account manager takes two days to review a submittal, no voice agent on earth is going to help you. Fix the review first. It’s cheaper.
And treat every vendor number in this category, ours included, as a hypothesis. McKinsey’s 2026 State of AI survey, covering 1,719 respondents across 97 countries, found only 37% of organisations could attribute any EBIT impact to AI, and just 6% qualified as high performers. Deloitte’s State of AI in the Enterprise 2026, across 3,235 leaders in 24 countries, found only 25% had moved most of their pilots into production.
Most of this fails. Measuring is how you find out which side you are on.
How to baseline your own number in four weeks
You cannot prove a reduction without a baseline, and almost nobody builds one. Four weeks, five timestamps, one control group.
The five timestamps to capture
- Requisition received. The email, portal notification or VMS release, not the moment somebody read it.
- Requisition assigned to a recruiter.
- First outbound contact attempt to the first candidate.
- First completed screening call.
- First candidate submitted to the client.
Four of those five already exist somewhere in Bullhorn, Avionté, JobDiva or Ceipal, usually in activity logging or the submission history object. The one that is normally missing is the first outbound attempt, because voicemails frequently do not get logged. Fix that first. Without it you cannot see the queue at all.
Run it against a control group or the number means nothing
Split incoming requisitions on the same desk, same client type, same period. Half worked the way you work today, half with the agent handling first contact and screening.
Not two different desks. Not this month against last month. Same conditions, split at random, or your number is a story rather than a measurement.
Watch submit-to-interview ratio at the same time
This is the guardrail, and it is the one people skip.
Time-to-submit is trivially easy to improve by submitting worse candidates faster. The only thing stopping that is watching submit-to-interview ratio in the same report. If your submits got faster and your interview conversion held, you gained something real. If conversion dropped, you just taught your client that your submissions need filtering.
What a faster submit is actually worth
Hours are not the unit your finance director thinks in, so convert them.
On a VMS panel, submission order correlates with interview slots, because hiring managers review in the order things arrive and stop reviewing once the shortlist looks full. Moving from day four to day one does not make your candidates better. It puts them in front of somebody while there is still room. The same race repeats once the shortlist closes, which is why interview scheduling is worth timing as carefully as the submit itself.
Run it on your own inputs. Take your annual VMS requisition volume, your current win rate, and your average gross margin per contract placement. Then ask what a few percentage points of win rate is worth against what the agent costs. That’s the entire business case, and it is arithmetic you can do on your own numbers rather than a promise anybody else makes. Our pricing is published rather than gated, so the cost side of that sum is not a phone call away.
Two things to keep honest while you do it. The staffing market is not doing the work for you: Staffing Industry Analysts forecast US staffing at $183.1 billion in 2026, growing 2.4%, then $187.0 billion in 2027, growing 2.2%. Growth is decelerating, so margin has to come out of operating cost rather than volume. And the labour market itself is soft, with the Bureau of Labor Statistics reporting 7.3 million job openings against 5.1 million hires in July 2026. Fewer reqs in the market makes winning the ones you get matter more, not less.
Getting started
Do the baseline first. Four weeks, five timestamps, one desk. Even if you never buy a voice agent from anybody, you will find out where your hours actually go, and most agencies are surprised by the answer.
Then, if the queue really is the bulk of your clock, test against it. We built Voicegun for staffing agencies around agency workflows rather than corporate talent acquisition, which means submission timestamps, VMS windows and native write-back to agency ATS platforms.
Book a demo and bring one real requisition. We will run the model on your numbers rather than ours.