Handshake's move to own labeling infrastructure
$1.1B/year Indeed for data labelers
The budget moved because data labeling buyers can swap vendors almost as easily as they swap staffing agencies. Labs already define the task, quality bar, and pay rate, so a new vendor mostly needs to recruit workers fast and plug into the same workflow. That is why shocks at Scale AI and Mercor translated into immediate share gains for Handshake, whose 18M student and graduate network let it stand up expert labor quickly while its campus SaaS business was shrinking.
-
This market behaves more like marketplace GMV than software ARR. Handshake estimates 60% to 70% of spend goes straight to contractors, leaving 25% to 40% gross margins on labeling work, which makes customer ownership fragile because the labor pool and the task spec matter more than product lock in.
-
The product is concrete. Labs send batches of model outputs or benchmark tasks, workers rank answers, write better ones, or flag failures, then the vendor returns labeled data. Mercor built around pre vetted experts and AI screening, Handshake around its existing PhD and graduate supply, but the core job for the buyer is still buying reliable human hours.
-
That is why Handshake is trying to climb the stack. Its January 2026 Cleanlab acquisition pushes it toward label auditing, evaluations, and RL environments, which are closer to the system a lab uses to test models every week, and therefore harder to rip out than a pool of contractors.
The next phase is a race to become the control layer above the labor pool. The winners will own evaluation workflows, quality monitoring, and recurring model improvement loops, not just contractor supply. If Handshake can turn a fast moving marketplace win into embedded testing infrastructure, its new revenue base should become much more durable than the recruiting business it is replacing.