Mercor's Shift to Expert Data Infrastructure

Mercor’s pivots show that the valuable asset is not generic recruiting software, it is a system for finding, testing, and routing scarce human judgment into AI training jobs. The company started with manual matching for engineers, then automated sourcing and screening with AI video interviews, and then moved further upmarket into doctors, lawyers, and scientists as labs began paying far more for expert feedback on reasoning tasks than for ordinary recruiting workflows.1 2 3 4

• The workflow changed from startup hiring to model training. Instead of helping one company fill one engineering seat, Mercor can now line up large pools of hourly experts who review model outputs, solve cases, and label reasoning quality for labs like OpenAI and Meta. That makes the marketplace look closer to expert data infrastructure than a classic recruiter.2 3
• This shift also changes unit economics. Mercor’s marketplace charges a subscription for platform access and a fee on full time hires, while the broader expert labeling category often books large top line volume that includes contractor payouts. That is why peers like Invisible and Handshake AI are discussed more like labor marketplaces than pure SaaS.2 3 5
• The closest comparables are no longer only Upwork or Turing. Mercor now sits in a newer cluster with Scale AI, Invisible, Prolific, and Handshake AI, all benefiting from labs needing credentialed people for math, law, science, and other high stakes domains where cheap generic raters are not enough.1 2 3

The next step is deeper verticalization. As labs and enterprises need more auditable expert data for regulated and specialized workflows, the winning platforms will look less like open job boards and more like managed systems for capturing professional judgment, measuring quality, and turning that work into repeatable training data and agent evaluation loops.1 3 4 5