ManpowerGroup Grows While US Temp Payrolls Shrink

AI-generated editorial illustration.

On October 15, ManpowerGroup reported $4.84 billion in third-quarter service revenue, up 4.4% from a year earlier. Two weeks before that release, the US Bureau of Labor Statistics had estimated 10,900 fewer temporary-help jobs in September than in August.

One number described a global company’s sales over three months; the other measured people on US temp-agency payrolls in one month. They cannot be netted against each other. They do put an operational question on the desk: how much of ManpowerGroup’s growth came from more assignments? How much came from share gain, service mix, price or a leaner delivery operation?

The company’s results release says Manpower led growth, especially in the United States. Its SEC-filed presentation also advances a transformation target: $200 million in permanent cost savings in 2028 and wider AI-enabled talent screening. Gross profit, however, was $767.3 million, slightly below the $768.9 million reported a year before. Revenue was higher; the gross-profit dollars were not.

The figures can coexist. A staffing company can win client volume while lower-margin work forms a larger share of sales. It can cut overhead even as gross margin softens. It can use AI in screening without showing that AI improved placement quality. None of these bridges is visible in the headline revenue number.

Two clocks on the same desk

The dates matter. ManpowerGroup’s third quarter ended September 30. The BLS September employment report, published October 2, estimated that US temporary-help employment moved from 2.4977 million in August to 2.4868 million in September, seasonally adjusted. It also revised prior months. That 10,900 decline is a month-to-month change in a national establishment survey, not a quarterly revenue trend for ManpowerGroup.

An industry survey adds a less pessimistic view. The American Staffing Association’s September staffing index release said its four-week moving average for the period ending September 13 was 5.8% above the comparable 2025 period, while its monthly index was 0.5% above August. The week containing Labor Day fell 1.4%. The association asks participating staffing firms about weekly employment; BLS samples employers for its payroll series. Different samples, seasonal treatment, time windows and comparison bases can point in different directions without one source being false.

Even the direction of the BLS month-to-month line deserves context. Its September report shows temporary-help employment down 10,000 in July, up 200 in August and down 10,900 in September, on the currently revised series. September was another soft month in a volatile series. A staffing firm’s contracts can book revenue for hours worked by people who started before September. A payroll count captures the number employed in the survey reference period. Neither detail cancels the warning; each narrows what it can say.

August’s broader US vacancy picture was subdued rather than collapsing. In its Job Openings and Labor Turnover Survey, BLS estimated 7.1 million job openings and 5.2 million hires, both little changed. Those economy-wide measures say little about the rate at which a particular staffing firm fills orders. They do explain why a client might buy flexibility while hesitating to commit to permanent hires.

ManpowerGroup’s own statement does not provide a reconciled US assignment count alongside its US revenue. The reported US business may grow through share capture even if industry employment slips. A large client may expand hours or buy a different service mix. Cross-border operations also sit inside the $4.84 billion headline. Anyone who treats the BLS monthly decline as proof that ManpowerGroup failed, or the global sales rise as proof that the US temp market recovered, skips the most important denominator.

The tension is familiar to a client with two open production lines. If one agency fills six shifts and another fills four, the first can grow without the factory hiring more contractors overall. Revenue can move between vendors without an industry hiring boom. Alternatively, one agency may invoice more because it supplied higher-paid specialists or more hours to the same headcount. A client would know which happened from its own purchase orders and time sheets. The public quarterly release does not show that level of detail.

Revenue rose; gross profit did not

The accounting bridge is plain this quarter. Service revenue increased from $4,634.4 million to $4,840.0 million. Cost of services rose from $3,865.5 million to $4,072.7 million. Gross profit slipped by $1.6 million to $767.3 million. The SEC presentation reports a 15.9% gross margin, down 70 basis points year over year. Extra sales did not produce extra gross-profit dollars in the quarter. That calculation says nothing yet about the economics of AI screening.

Lower selling and administrative expense did help. SG&A fell from $702.3 million to $685.0 million, and operating profit rose from $66.6 million to $82.3 million. The reported operating margin was 1.7%; adjusted EBITA margin, a different management measure, was 2.2%. These figures describe a company getting more operating profit from a thinner gross margin by controlling costs below gross profit. They do not isolate a technology return, and the adjusted measure should not be substituted for the reported margin.

Country and brand mix complicate the aggregate. In the results table, US revenue from services reached $728.8 million, up from $690.8 million. France, a much larger market in this reporting structure, fell from $1,173.5 million to $1,132.9 million as reported. Management’s prepared comments attribute part of the gross-margin pressure to softer permanent recruitment in France and lower Right Management outplacement volume. They also describe stronger US Manpower demand in consumer goods, food manufacturing, logistics and specialized technical roles tied to automation and reshoring. None of those descriptions supplies a company-wide placement count by product or an AI versus non-AI cohort.

Last quarter is a useful check against an easy narrative. In its July second-quarter release, ManpowerGroup reported roughly $4.9 billion in revenue and said gross-profit growth joined lower SG&A to improve profitability. The third quarter is different: revenue grew year over year, but gross-profit dollars slipped. One good quarter of expense control should not be stretched into a claim that each new unit of sales has become more profitable. The next mix disclosure matters as much as the next revenue number.

The problem is not limited to the income statement. A staffing branch can look productive when its recruiters handle more requisitions, yet deteriorate if cancellation, no-show or early-attrition rates rise. A high-value technology engagement can lift revenue per consultant while tying up senior delivery staff for longer than the contract assumed. Neither effect appears in the consolidated gross-profit change as a clean, separately labeled AI line. That is why operating profit is necessary evidence of management progress but insufficient evidence of any particular tool’s return.

For an investor, three margins are easy to collapse into one. The wage-and-benefit spread on a temporary assignment sits in gross profit. Branch rent, central technology, sales teams and other operating costs enter later. A digital service contract carries its own delivery labor and implementation timing. Stronger operating profit can coexist with a lower gross margin because the company spends less below that line. An efficient branch screen could produce the same pattern; so could an unrelated cost action. The quarterly filing reports the aggregate, not a branch-by-branch causal study.

For a client, a cheaper agency process does not automatically mean a lower invoice. The agency may retain the savings to fund a better service level or repair its own economics. Or it may price more aggressively to win an account, exchanging some margin for volume. Both can be rational. The negotiation should ask for reliable starts, worker retention and transparent service levels before assuming that a faster screen warrants an immediate fee reduction. The client’s bill is a different ledger from the vendor’s SG&A.

The commercial claim is plausible: Manpower can gain US accounts even when a national temp series is soft. The cost story has a firmer footing in the reported SG&A decline. Yet the release does not allocate operating-profit improvement among recruiter workflow changes, ordinary cost actions, client mix, price and timing. An AI productivity dividend remains unquantified.

The $200 million bridge to 2028

ManpowerGroup says its strategic transformation is on track for $200 million in permanent cost savings in 2028. The filed presentation says the program expanded to the UK in the third quarter and that AI-enabled talent screening is widening. Management also guided to fourth-quarter gross margin of 15.8% to 16.0% and operating profit margin of 2.1% to 2.3%. A target and guidance are forward-looking; neither is a booked saving.

For a staffing network, a savings program has several possible locations. Shared back-office tasks can move into common systems. Branch capacity can be adjusted to order flow. Sourcing and screening may require fewer repetitive steps. A client can buy more specialized work at a different price. These mechanisms have different effects on worker service, recruiter workload and client retention. A lower SG&A line is real, but it does not say which mechanism contributed how much.

The transformation began before this quarter. In first-quarter prepared remarks, CFO Jack McGinnis said IT and finance savings would build during 2026 toward roughly a quarter of the 2028 target. Front-office changes, including standardized processes and automation, would come later as the program expanded. That sequence rules out a simple reading in which every dollar of the current SG&A decline came from AI screeners in branches. The savings target spans systems, geography and years; the Q3 result covers one quarter.

Reaching the target has a cost. The earnings release says restructuring and strategic transformation program costs affected the reported quarter. A shareholder should ask how much cash and expense it takes to produce recurring savings, when those savings appear, and whether client retention or recruiter service changes. Adjusted EPS excludes specified items. The exclusion helps compare periods, but it does not erase the implementation bill. Permanent savings count only after that bill and any lost service quality are considered.

The gross-margin line makes the distinction consequential. Screening may save a recruiter’s time while temporary workers’ wages, payroll taxes, benefits and insurance still move through cost of services. A service mix shift can bring in more gross revenue with fewer gross-profit cents per dollar. At the same time, a cheaper initial screen can create more candidate contacts and more exceptions that a human must resolve. Without a disclosed cost-to-fill or assignment-level margin series, an apparent productivity improvement could be a timing effect or a transfer of work to another team.

Becky Frankiewicz will inherit that measurement problem. The board named her CEO effective January 1, 2027, with Jonas Prising moving to executive chair. Frankiewicz has led the commercial strategy and, according to the company, integration of AI across its brands. Her first annual comparisons will have to separate a market rebound from the savings program and from the contribution of new digital services. Otherwise a $200 million target can look precise while the route to it remains hard to audit.

Screening is an input, not a filled shift

The company has described expansion of AI-enabled talent screening, but its Q3 presentation does not show a controlled comparison of screened candidates and completed assignments. That missing conversion matters. The unit a client pays for may be a worker who starts, a shift that gets covered, a specialist who stays through a project, or a managed service result. A screen is earlier in the chain.

Imagine a branch handling a food-manufacturing order for 80 workers across two shifts. Faster candidate ranking might shorten the first call list. The branch still has to confirm location, transport, wage expectations, eligibility, safety requirements, shift availability and a supervisor’s acceptance. A name moved into a shortlist is not a shift worked. If the automated shortlist omits people who could actually make the shift, recruiter time may fall while fill rate deteriorates. If it catches the right people quickly, the client may get more reliable coverage with less overtime. Public disclosures do not choose between those cases.

AI-generated editorial illustration. The three paths separate completed assignments, exception work and service economics.

A useful test would follow the same requisition cohort from first contact through accepted offer, start, completed assignment and repeat order. It would measure candidate response time and opt-out as well as speed. It would show recruiter and operations minutes consumed per completed placement, including exceptions and manual review. And it would track gross profit after the direct costs of service. Those numbers are not promised as public company metrics here; they are the internal evidence required to support a claim that automation changed the unit economics rather than merely the front of the funnel.

Candidate experience belongs in that calculation. A worker who receives an instant screening message but waits days for a clear assignment decision has not necessarily been served better. A recruiter who must correct a bad match or explain an opaque rejection has not been freed from work. The fair comparison is a qualified, informed and completed handoff at a known cost, not the number of prompts a system can send.

The worker’s denominator differs again. A night-shift applicant may value a fast answer more than a polished assessment summary. A person with irregular transport needs a realistic location and start-time check before spending an hour on onboarding. Someone rejected after a screen needs to know whether another assignment is still open. Those are service obligations, not decoration around a placement engine. An agency could improve fill rate by contacting more people while making each person’s experience worse. It could also use a better first contact to reduce wasted trips and abandoned starts. The company has not published the cohort evidence needed to judge which outcome its wider screening rollout produces.

The client may also resist a clean experiment. An urgent plant order is often given to the branch most likely to fill it, rather than randomly assigned to an AI-assisted process. Pay rates and travel distances move while the trial runs. A useful comparison has to document those differences and show sensitivity to them. Otherwise the system can take credit for easier orders, or be blamed for the hardest ones. That objection is strong enough to complicate the audit, not to excuse having no audit at all.

There is a second selection problem: who reaches the screen at all. An agency might invite only candidates who already meet availability rules, improving the apparent screen-to-start conversion without increasing the number of people hired. Another might invite a broader group, making conversion look worse while giving previously overlooked workers a chance. A ratio without the size and composition of its entry pool rewards whichever intake rule makes the dashboard look best. A defensible report carries the invitation denominator, not merely the completion count.

The worker should have a path back to a person when the automated step cannot resolve a real constraint. A missed call after a shift change, a disability accommodation request or an incorrect availability flag can turn an apparently quick process into a dead end. A company may already handle these cases well; the public Q3 materials simply do not say. Counting exceptions alongside placements is therefore part of testing the expansion, not an accusation that a specific system has failed.

A services business beyond temp staffing

ManpowerGroup is not only a temp agency. Its Experis brand sells professional staffing and technology services; Talent Solutions includes other workforce offerings. The SEC presentation says the Manpower brand grew 7% on an organic constant-currency basis, Talent Solutions grew 1%, and Experis was flat globally after a decline in the prior quarter. US Experis returned to growth. These are brand revenue trends, not proof that any specific AI product caused the change.

In July, Experis US launched ExcelerateWorkflow, a service built with IBM watsonx Orchestrate to help enterprise clients deploy AI-powered workflows with human oversight. The release said initial US client deployments were underway and global expansion was planned. It did not disclose product revenue, contract values, margins or measured customer outcomes. That matters when assessing the staffing group’s future mix: selling AI implementation work is a different business from using AI to source temporary workers, even when both appear under the same corporate earnings headline.

Kye Mitchell, the president of Experis US, framed the offering around connecting technology to a client’s systems and people. A client may pay for that implementation because it lacks the internal specialists to make an AI pilot useful. The same client may reduce a separate staffing order if a workflow removes repetitive work. ManpowerGroup can participate in both transactions. It would be misleading to infer one-for-one replacement from an individual product launch, but it would be equally misleading to treat every dollar of future technology services as evidence of more jobs placed.

The buyer’s questions differ. For temporary labor: can the firm cover a role safely and reliably at a fair markup? For AI implementation: does a governed workflow reduce work or improve service without creating a larger support obligation? A group that sells both may deepen its client relationship. Its consolidated revenue still does not reveal how many hours of worker employment each dollar represents.

Consider a hypothetical operations leader buying both services. She needs fifty temporary warehouse workers for a seasonal peak and a separate Experis team to automate invoice exceptions. The first purchase is judged by starts, attendance, safety and replacement speed. The second is judged by whether invoices clear with fewer staff minutes and acceptable error rates after the implementation team leaves. A combined contract may be convenient, but it does not justify one blended outcome metric. Each promise has a different unit, risk and renewal decision.

It also changes the competitive story. A rival staffing firm could have weaker software and still win by keeping difficult shifts filled. A specialist AI integrator could build a better workflow without operating a labor branch. ManpowerGroup’s cross-brand offer may be valuable precisely because it can see both workforce and workflow needs; the Q3 figures do not establish that cross-selling occurred, how much it earned, or whether the same clients bought both. The investor case for that combination is an inference to be tested against future segment disclosure.

A ledger a client can actually request

A procurement team evaluating an AI-assisted staffing program can ask for a small, paired operating report before expanding it. The contract should specify a comparable baseline: similar requisitions, locations, shifts, wage bands and time periods. It should separate cases touched by AI from cases that were not, with changes in order mix called out. A before-and-after average without those fields invites the vendor to confuse a better market with a better process.

Line in the operating report	Why it belongs in the decision
Requisitions received, accepted and canceled	Keeps demand and vendor selection separate from fill performance.
Candidates contacted, screened and human-reviewed	Shows where automation changes the funnel and where people still intervene.
Qualified submissions, starts and completed assignments	Connects a screening metric to actual client coverage.
Time to first response and time to start	Distinguishes a faster message from a faster staffed shift.
Early attrition, safety incidents and client reorders	Tests whether speed has shifted costs downstream.
Recruiter, coordinator and exception minutes per completed assignment	Includes labor moved out of the visible screening step.
Gross profit per completed assignment and total client spend	Tests whether a lower handling cost produces a durable economic gain.

There are limits to such a report. Worker privacy and local employment law constrain what candidate-level data a client should receive. Assignment types differ. A large client’s most urgent orders may land in the AI-assisted cohort first. The report should therefore use aggregated, comparable cohorts and explain exclusions. It is an operational audit, not a demand for individual candidates’ screening records or a claim that automation can be evaluated with one universal score.

The report also gives a fair hearing to ManpowerGroup’s case. If AI screening helps recruiters reach willing, available workers faster and the agency converts that into reliable starts, the client sees a benefit even if a national payroll series declines. If margins improve because the company serves the same orders with fewer back-office steps, investors may see it before gross margin rises. Those are testable paths. A sales line by itself leaves them unresolved.

Numbers still missing

The next BLS temporary-help estimate will still be a national monthly sample. The next ManpowerGroup quarter will still be a global company report. Neither should be asked to settle the other’s question. The more revealing disclosures would be US Manpower assignment volume or billed hours, a bridge between staffing and services mix, and measured savings by workstream with the costs of implementation and exceptions included.

The visible record supports a narrow conclusion. ManpowerGroup sold more services, particularly in US Manpower, and earned more operating profit. Its gross-profit dollars did not grow. US temporary-help payrolls fell in September, while the ASA’s differently measured series showed year-over-year growth. AI screening expanded and a $200 million savings target remained in place; neither disclosure gives a placement or customer-outcome denominator. A branch that fills an additional shift can prove something the headline revenue number cannot. Until that bridge appears, the recovery and the automation story remain separate claims.