Where staffing firms actually lose placement hours

Bullhorn’s GRID survey is the industry’s flagship annual research, and its 2025 edition reported that two thirds of staffing firms had purchased an AI solution, built one, or were experimenting with generative AI. That number gets quoted constantly.

Read the decomposition:

So “two thirds of the industry is doing AI” means one firm in seven has bought anything, and half the industry is trying ChatGPT.

And almost all of it went to the same place. GRID 2025 found 45% experimenting with AI to sort candidate resumes and submissions.

Screening. The part a recruiter already enjoys.

The vendor is telling you where it stopped

The 2026 edition of the same survey, across nearly 2,300 recruitment professionals, contains three consecutive sentences that are worth more than the rest of the report:

“54% of firms report that they have automation in place for search. That percentage is even lower for middle office functions like payroll and billing. Fewer than half of firms are using AI for any individual recruitment function.”

— Bullhorn, GRID 2026 Industry Trends Report

That is Bullhorn. The dominant staffing ATS vendor, with its own AI product line, publishing that automation peaks at candidate search and thins out toward the back office.

Kevin O’Neill, Technology Research Director at Staffing Industry Analysts, dated the shift in August 2026:

“Eighteen months ago, staffing technology vendors told SIA there was ‘still work to be done’ before AI would find its place in the middle and back office. That position has changed decisively.”

— Kevin O’Neill, Technology Research Director, Staffing Industry Analysts

The tooling for the back office arrived eighteen months after the tooling for screening. Which is exactly why every firm’s AI programme stopped where it stopped.

The hours are in the copy, not the CVs

Here is what a recruiter’s week actually contains that nobody has counted.

One job order gets rewritten for a dozen client portals, each with its own submittal format, rate card structure and character limits. A candidate summary gets reformatted three times for three clients who want the same information arranged differently. Timesheets arrive from paper, from client time clocks and from a VMS, and get reconciled by hand against an invoice and a payroll run, with the exceptions surfacing after cutoff rather than before.

None of that is screening. All of it is writing, and it happens on the clock of the person whose only revenue-generating activity is placement.

There is one piece of empirical support for that ordering, and it is a good one. StaffingHub’s 2026 State of Staffing survey looked at which AI use cases correlated with revenue growth.

The strongest were job description generation at 2.7x and reporting and analytics at 2.7x. Candidate qualification came in at 2.2x.

The industry’s own survey puts job order copy at the top, ahead of screening. It is a small sample and a correlation, and it is still the only ranking anyone has published.

“What we’re trying to do is redesign everything to be AI-native, instead of just trying to bolt AI on. I don’t think that’s the effective way.”

— Matt Linneman, VP of Key Accounts and Recruiting, CEI

What four hours a week is worth on your floor

Now the arithmetic, and you can substitute your own inputs as we go.

Start with what a desk is worth. ASGN’s 2025 annual filing reports roughly 2,800 internal employees and $1.15 billion in gross profit, which is about $410,000 of gross profit riding on each internal head — and that figure is diluted by every corporate and back-office employee in the company, so the number per producer is higher. A fractional-CFO practice that benchmarks staffing firms puts the median gross profit per producer at around $300,000, with the top quartile above $400,000. Two unrelated methods, the same neighbourhood.

Take the conservative one. Twenty-five producers at $300,000 is $7.5 million of gross profit.

A producer works roughly 2,160 hours a year, at 45 hours across 48 weeks. Now put in the number nobody measures: hours per week lost to rewriting job orders for client portals, reformatting submittals, and chasing timesheet and invoice exceptions.

Use four hours. Deliberately low.

Four hours a week is 192 hours a year, which is 8.9% of a producer’s capacity. Across twenty-five desks that is $667,000 of gross profit riding on time that produces nothing.

Then halve it, and say out loud that you are halving it. A recruiter handed back an afternoon does not automatically bill it. At fifty percent conversion, roughly $333,000 a year of gross profit in a twenty-five-desk firm, from four hours a week.

That is about thirty placements. Run it on your own numbers:

Producers × gross profit per producer × (hours per week × 48 ÷ 2,160) × 0.5

If you think four hours is low for your desks, you are probably right, and the number moves fast.

Set the result against what it costs to test it. A first step on one desk starts at $15,000 and reports in six weeks. At the conservative end of the arithmetic above, that is roughly two placements.

Nobody measures this, which is why it survives

Two claims circulate in this industry that you should stop repeating.

The first is that recruiters spend 80% of their time on admin. It has no traceable source. Every page carrying it is a software vendor with no survey instrument, no sample and no method behind it.

The second is subtler and more common. Bullhorn’s GRID 2025 published a set of figures showing 4.5 hours per week on search and match, 3.6 on screening, 3.6 on administrative tasks, totalling 17 hours a week. Those are a projection of what AI could save, not a measurement of what recruiters currently lose. The figure now circulates in staffing marketing as though it were an observation. It is not.

Cost-of-vacancy calculators belong in the same bin. Every one traces to an agency or RPO blog, and the formula is salary divided by working days multiplied by days open, which is an arithmetic identity dressed up as a finding.

What is genuinely missing is more interesting than what is wrong. There is no published figure anywhere for how long it takes to rewrite a job order for a client portal, format a submittal, or reconcile one timesheet-to-invoice cycle. GRID 2026 asks about screening in granular detail across 2,300 respondents. It addresses the back office once, in a subordinate clause.

The industry’s flagship survey does not ask the question. Which is a reasonable explanation for why the hours have never gone anywhere.

The VMS problem is structural, not a preference

The reason job order copy multiplies is not disorganisation. It is the channel.

Staffing Industry Analysts put managed service provider spend under management at $226 billion globally in 2024, with the Americas at 56% of it, and 59% of new MSP contracts going to organisations adopting their first external MSP. The channel is still expanding.

And MSPs run more than one platform.

In SIA’s own data, 80% of MSP providers use SAP Fieldglass, 77% use Beeline, 47% use Workday VNDLY, and 21% of clients procure the technology separately from the MSP. Those figures sum to well over a hundred percent because a single MSP typically runs several.

So a staffing firm supplying into five MSP-managed programmes is not dealing with five clients. It is dealing with five programmes that between them run three or four different VMS platforms, each with its own submittal format, rate card structure and timesheet approval architecture.

Meanwhile StaffingHub found the average agency running 5.5 separate software platforms, with only 25% having most or all of them integrated.

That is the reconciliation tax, and it is charged to a recruiter’s week.

You cannot grow your way out of this

SIA projects the US staffing market at $180.2 billion in 2026, growing 1%. That follows a 3% decline in 2025, a 12% decline in 2024, and a 14% decline in 2023. The market is still smaller than it was before the pandemic.

Three down years and a 1% recovery means the hours have to come from inside the business, because they are not coming from the market.

At the same time the bar moved. In GRID’s 2024 data, the growth correlation sat with firms placing in under twenty days. In the 2025 data it had moved to under ten days, with 56% of firms growing more than 25% placing inside that window, against 34% of revenue-declining firms sitting at ten to nineteen days.

The threshold for “fast” halved in a single year. The hours to clear it have to come from somewhere.

“AI won’t take your job, it will help you get better at it, but what will take your job is another recruiter at a competitor that’s using AI and automation really well.”

— Sam Porter, IT Director, Morgan Hunt

This is the same pattern visible across every industry that runs the physical world: the tools land in the function that is easiest to demo, and the money stays in the function nobody measures.

Where the first step goes

If the hours are in the writing and the reconciliation, that is where a first step belongs.

Recruiting and delivery. Job order copy drafted for a dozen client portals, candidate screening summarised into consistent submittal formats, inbound resume floods triaged before they reach the ATS. This runs alongside Bullhorn, Beeline, Ceipal, Avionte, whatever the desk already runs on. It is one desk, measured, from $15,000.

The six weeks, concretely: a week timing how one desk currently rewrites job orders and formats submittals, two weeks configuring agents against your own ATS records and your own past submittals in each client’s format, three weeks of that desk working live orders with it, and a written comparison at the end. Your ATS and every client VMS stay exactly where they are.

Back-office reconciliation. The heavier version: an embedded pod rebuilds timesheet to invoice to payroll end to end, ingesting hours from paper, client time clocks and the VMS, and flagging exceptions before cutoff instead of after. That runs against live data with a go or no-go gate before anything scales.

The full picture of what AI-native staffing looks like runs from the front desk to the back office. No firm should start with all of it.

The number to go and find

Everything above is somebody else’s data. Here is yours, and one recruiter can answer it in an afternoon.

Take one job order that went to more than three clients last quarter. Count how many times the copy was rewritten. Time one rewrite. Multiply by your open orders.

Then ask a recruiter how many hours last week went to submittal formatting and timesheet exceptions. The answer will be higher than four.