The industry is shrinking, and officially so. The Federal Employment Agency counted 669,000 temp workers in June 2025, the lowest level since 2010 and 38 percent below the late-2017 peak. Market revenue fell for the second consecutive year to €31.9 billion per Lünendonk, Creditreform counted 120 staffing insolvencies in 2024 plus 63 in the first half of 2025, and reports 27.8 percent of firms with an equity ratio below ten percent. Austria's official statistics show leased workers down 10.2 percent in a single year. Recruitment consulting is shrinking alongside: 2,175 mostly tiny firms, €2.78 billion forecast revenue, with the industry association BDU itself noting that clients' AI-assisted candidate search dampens demand for consultants.
At EBIT margins that industry portals consistently put at 2 to 6 percent, no official statistic exists, honestly, the consequence is stark: whoever loses recruiter hours to manual work is not losing growth but substance.
The work scales with churn, not headcount
The most uncomfortable number in the agency data is not the decline but the turnover: 789,000 new employment contracts in twelve months against a stock of roughly 660,000 workers. A staffing firm runs the full intake pipeline, CV capture, document checks, staffing-law paperwork, equal-pay calculation, onboarding, more than once per worker per year on average. In Austria the average assignment now lasts just 73 days. Add the process reality from 2.5 million real applications in the onlyfy application report: one hire takes on average 70 days, 19 applications and 4 interviews, while the agency-registered vacancy duration has climbed to 169 days. Three quarters of applicants expect an interview invitation within two weeks, and per a December 2025 KOFA study, 14.7 percent of companies have seen signed contracts simply not honored on day one. In this business, speed is not a virtue; it is the product.
One more finding quietly eroding the screening logic: per a June 2026 softgarden study, 43.2 percent of applicants write their cover letters with AI, three times the 2023 share. The signal recruiters read by hand is increasingly machine-generated.
The trap: the most obvious AI use is the legally worst one
Annex III of the EU AI Act declares high-risk those AI systems intended, verbatim, "to analyse and filter job applications, and to evaluate candidates". CV screening, ranking, automated shortlisting, exactly what US vendors market most aggressively, is named explicitly. After the Digital Omnibus the obligations apply from 2 December 2027, and they were deferred, not diluted: competent, named human oversight, logging, informing affected people. Whoever builds or white-labels their own matching model risks full provider status including conformity assessment. And Germany's anti-discrimination law shifts the burden of proof in § 22 AGG: a rejected candidate needs only indications, and then the agency must prove its opaque model did not discriminate.
Two numbers vendors will show you anyway do not survive scrutiny: the "6 seconds per CV" comes from a 2012 eye-tracking study with 30 participants, and the "75 percent of applications die in the ATS" is a misreading of a Harvard study that never measured any such thing.
The inverted funnel: automate everything except the judgment
For a staffing or recruiting firm of 10 to 100 people, the gains sit precisely in the pipelines the AI Act does not regulate and the daily business devours:
- Intake and qualification: applications and client orders captured in structure, documents completed, follow-up questions asked automatically, in the firm's tone and with sign-off. That is data collection, not selection.
- Reactivating your own pool: a churn of 789,000 contracts a year means every agency sits on a treasure of already-vetted, known workers. Systematic, occasion-based re-engagement is the cheapest placement in the market and entirely unregulated.
- Scheduling and communication: interview scheduling against the two-week expectation, status messages against ghosting in both directions. Half of candidate drop-outs get attributed to slow communication in industry surveys, read that as a vendor number, but the direction is plausible.
- Above all: the compliance paperwork. The staffing license, maximum assignment duration, equal pay after nine months, sector surcharges, in Austria the reference-surcharge logic and the training-fund levy. These documents decide the license and thus the firm's existence; drafting them and monitoring their deadlines is the safest and most valuable AI deployment in the entire industry.
Where AI does touch selection, four rules apply: bought rather than built, deployed as documented decision support rather than decider, with a named human per shortlist, logged and disclosed. That is exactly the draft-plus-sign-off principle we build everywhere, here simply with the force of law behind it.
The survival math of a shrinking market is simple: what remains are the agencies whose recruiters spend their hours on the two things machines may not legally own here, the matching judgment and the client relationship. Whether your firm has an automation case, the 60-second check will indicate; what projects cost is in the pricing overview, with sources.
As of 10 July 2026, not legal advice. Sources: Federal Employment Agency Zeitarbeit report (January 2026), Lünendonk list 2025, Creditreform (October 2025), Statistik Austria staffing survey (2025), BDU Facts & Figures 2025, onlyfy/XING application report 2025, KOFA study 4/2025, softgarden "KI trifft Recruiting" (June 2026), EU AI Act Annex III 4(a) and Art. 26 (post Digital Omnibus, June 2026), § 22 AGG, German and Austrian staffing acts. Debunks: the "6-second CV" (2012, n=30) and "75 percent die in the ATS" (misreading).