Dispatching to DevelopWorker Agency and Workforce Development in Gig Platforms

Many gig platforms allocate work by controlling which opportunities workers see first. A platform can give selected workers priority access to an order, but independent workers decide whether to accept, and the assignments that result may improve future performance through learning by doing. We study how a platform should manage this visibility when workers differ in current capability, long-run potential, learning speed, and willingness to accept work. In a finite-horizon model, the platform exposes each job to a limited set of workers, observes acceptance decisions, and assigns the job to an acceptor. The exposure decision trades off current reliability against the possibility of developing workers whose value is not yet fully observed. We first show that the fully informed dynamic optimum is too strong as an online-learning benchmark because it knows which inexperienced workers are worth developing before the platform could have generated that information. We then introduce a finite-horizon planning oracle and develop optimistic exposure policies, including a reliability-constrained version that limits unfilled jobs. Using operational data from a large U.S. retail-delivery platform, we document selective exposure, bounded performance improvement, heterogeneous development headroom, and participation patterns that are not uniformly increasing in capability. Calibrated policy comparisons show that development through dispatch is valuable only under the right conditions: broad training is costly, aggressive optimism need not improve reward when many workers enter with substantial capability, and reliability constraints make the service-development trade-off explicit. The results suggest that dispatch systems should be evaluated not only by how well they fill current demand, but also by how selectively they build the capacity needed for future demand.

From SSRN, via Crossref