Online real estate platforms make property listings widely accessible, yet buyers’ effective search opportunities may still be shaped by agents’ informal collaboration networks. Using detailed records of online browsing, offline property showings, and transactions in the Beijing housing market, we recover agents’ collaboration networks from their historical interactions using an unsupervised machine learning method. We find that after contacting an agent, buyers shift their online search toward locations with greater coverage by their agent’s network, and network-linked properties are more likely to receive offline visits, even conditional on prior awareness and local market characteristics. We develop and estimate a structural model of buyer awareness, search and purchase that distinguishes two channels through which agent networks operate: shaping which listings buyers become aware of, and steering which known listings they visit. Counterfactuals separately eliminate network-based advantages in information exposure and offline visit recommendations to quantify their effects on search depth, match quality and buyer welfare.

Star employees are among a firm’s scarcest resources, making it important to understand how they create value and where firms should deploy them. While prior research emphasizes supply-side spillovers such as knowledge transfer, peer learning, and collaboration, we study a different mechanism arising when customer demand attaches to individual employees. Using intra-firm reassignments of agents across shops of a large real estate brokerage, we find that shop performance rises immediately and persistently when a star arrives and reverses when she departs. Star arrival sharply expands the receiving shop’s customer pool. Yet the star does not increase her own service volume; instead, coworkers absorb the additional demand, lifting their output by 11.8 percent. Tracing customer flows across shops, we find that about two-fifths of the receiving shop’s customer gain is redirected from the star’s origin shop, while most of the inflow is net new demand the firm would not otherwise have captured. But shop capacity is bounded: as coworkers take on more buyers, conversion does not improve and buyers search longer, with the strain sharpest where the shop is already congested. The same star creates more value in shops with room to serve additional customers and the ability to convert them. These findings reframe star deployment from fostering post-arrival coworker interaction to matching scarce demand-drawing talent to the units best positioned to capitalize on the demand it attracts.