Beyond Relevance: A Demand Balancer Model for Rental Platforms with Single-Unit Inventory

House rental marketplace platforms face unique challenges due to their single-unit inventory nature, where each property is distinct and can only be rented to one tenant. Traditional ranking systems typically optimize for user-item fit [1--5]. For the rental market, this optimization creates demand bottlenecks by continuously directing users to already popular properties. This approach results in high-competition scenarios where multiple users are drawn to the same high-relevance properties, generating excessive competing offers. This can lead to user frustration, as only one individual can ultimately secure the property, leaving others dissatisfied despite their high compatibility with the listing. At the same time, given the nature of property technology businesses being inherently supply-constrained, demand concentration negatively impacts landlords as well, creating challenges for those struggling to rent out their properties, facing longer vacancy periods and increasing the risk of renting elsewhere. To redistribute demand across our house rental marketplace, our solution incorporates the likelihood of successful conversion based on historical user-house interaction signals, including visits, offers, and others. By dynamically adjusting property visibility based on predicted rental probability, we effectively redistribute user attention to low demand properties while minimizing losses in relevance. The implementation of this system in a large-scale rental marketplace through an online controlled experiment resulted in 4% increase in unique houses receiving offers from users and a 3% improvement in contract conversion rates. These results suggest that incorporating availability predictions into ranking systems can lead to more efficient marketplace dynamics while maintaining user satisfaction, without impacting user engagement. Our approach provides a framework for balancing demand in marketplaces with unique inventory constraints.