How AI Revolutionizes Recruiting

The main pricing model for over three decades in external recruiting has been the contingency fee. Agencies are paid a percentage of the candidate's first year base salary, usually 15 to 20 percent (Society for Human Resource Management [SHRM], 2023). These fees are only paid once the candidate is hired. For a software engineer earning $150,000, a single placement could net the agency $22,500 to $37,500. The model became standard because it aligned incentives. Agencies were not paid for trying, only for delivering a candidate, and the fee reflected the underlying labor cost of the work. In other words, the better the candidate suited the position, the more the agency was paid.

The contingency model is a bet on labor productivity. If an agency places one engineer per recruiter each month at a 20% fee on a $120,000 salary, they would generate $24,000 in revenue per recruiter. After commission splits, overhead, and database subscriptions, gross margins on a generalist desk historically ran between 25% and 40% (Bullhorn, 2024). This works because each placement absorbed many hours of human screening, calling, and qualifying. This is the time the hiring company either could not or did not want to spend. They believe that by outsourcing this service, the company will benefit in the long run.

Recruiters were paid for the information that they had, information that companies that needed workers did not have. Until the mid 2010s, labor markets were centered around something called information asymmetry. This is the idea that in a transaction, one party has much more information than the other (Akerlof, 1970). These companies needed talent that existed in the job market. However, the cost of finding the talent, reaching out, and getting them to interview was too large for many HR departments. This creates a service gap. Recruiting businesses filled this gap by delivering their services. They created and maintained rolodexes, applicant tracking systems, and LinkedIn searches of candidates. This prior knowledge was a large part of what companies paid recruiting services for. As long as there was information asymmetry, the recruiters could take advantage of the service gap. This is what allowed them to take large percentage fees. When this gap closes, and the companies have the same information as the recruiters, there is no longer a need for their services.

The recruiter industry secret pre-AI was that most of the daily work was somewhat low-skilled. Much of a sorcerer's daily work was scrolling through LinkedIn, reviewing anywhere from 200 to 400 profiles in a day. These profiles could lead to about 20 possible hires, 3 interviews, and 1 job placement (LinkedIn Talent Solutions, 2022). Estimates from Bersin by Deloitte (2021) put the average time-to-fill in the United States at 36 days, with the majority of that time spent in candidate identification and screening rather than in interviewing or negotiation.

This is the filtering bottleneck. The primary input cost in the old model was human attention applied to a digital haystack. Salaries for sourcers in U.S. metropolitan areas typically ran $50,000 to $75,000, and a productive desk required one to two sourcers feeding each senior recruiter (Built In, 2023). Once that cost is recognized as labor, the entire economic structure of the industry becomes legible. The contingency fee was effectively a markup on hours spent looking, not hours spent judging.

Once labor is identified as the dominant cost, the strategic implication is unavoidable. Any technology that meaningfully reduces the hours required to find and qualify a candidate will do one of two things. In a market with weak competition, it will shorten those hours into the same fee structure and increase profit margins, because the agency keeps the price but lowers the cost. In a market with strong competition, or one in which the buyers (employers) gain leverage, it will collapse the price, because competitors can undercut a 20% fee with a 10% fee and still earn an acceptable return.

The recruiting industry has historically been low-barrier and highly fragmented. There are tens of thousands of staffing firms in the United States alone (American Staffing Association, 2024). In fragmented industries, cost-reducing technology tends to deflate prices rather than fatten margins, because there is no oligopoly to enforce pricing discipline. This is the central economic premise that the rest of this paper builds on. AI is most dangerous to the recruiting industry not because it is unfamiliar, but because it attacks the exact cost line, human filtering labor, that justified the price line.

The recruiting market is shaped more by the platform that owns the candidate graph than by the agencies doing the placement. In this case, the platform is LinkedIn. They are an example of Thompson’s (2015) aggregation theory. This is the idea that aggregators win by owning demand and setting the terms. They do not win by producing supply. Essentially, everybody is already on the platform. White-collar employees all have profiles, and they all look for jobs through LinkedIn. Recruiters cannot afford not to use LinkedIn. They cannot negotiate with the platform. They need to pay for LinkedIn or lose access to the data pool. LinkedIn makes its money mainly through Recruiter Seats, which in 2024 ran about $13,000 per seat per year (LinkedIn Talent Solutions, 2024). This is a fixed cost that agencies pay before they've placed anyone.

LinkedIn is also two things at once: a supplier with enormous bargaining power, and a potential entrant into the recruiting business itself. The 2026 launch of Hiring Assistant made that second part explicit: LinkedIn isn't content to just charge rent for access to candidates anymore; it's now competing with its own customers for the placements (LinkedIn, 2026).

The most important shift of the LinkedIn era is a setup for the AI argument that follows. Search costs collapsed. Finding a qualified candidate is now a query, not a week of cold calls. Confirming that someone can actually do the work, that the resume dates are real, and that the references aren't friends pretending to be former managers is just as hard as it ever was. Industry surveys consistently find 30–40% of resumes contain material misrepresentations, roughly flat for two decades (HireRight, 2023).

That asymmetry is what creates space for both the agentic AI threat in Section 4 and the high-touch survival path in Section 5. When search is free but verification is hard, the value in the industry migrates from finding people to vouching for them. Whoever can credibly verify is the one who keeps getting paid. The open question is whether algorithms or humans will own that step next, and this will decide who wins and who loses over the next three years.