Heterogeneous Complementarity and Team Design: The Case of Real Estate Agents

Yan Xu | Mantian Hu | Junhong Chu | Andrew T. Ching

Jan 23, 2026

Key Takeaways

Research Question: Workers often possess characteristics that are critical for teamwork but are not directly observed by managers, for example, work style. These unobserved traits can be captured by latent types. This paper seeks to answer several key questions about team design: Do different latent types of workers complement one another to different degrees? If so, which types of workers are better team players? And how can firms leverage these heterogeneous complementarities to design more productive?
Data and Method: The study uses seven years of team-assignment and team-performance data from a major Chinese real estate brokerage. The authors developed a teamwork model—building on Bonhomme (2021) and the stochastic model— to flexibly estimate workers’ latent types and the heterogeneous complementarities between them. The model quantifies pairwise complementarities without imposing functional-form restrictions and provides team-assignment policies that can significantly improve overall performance.
Findings
- Team success depends on complementarity, and team design using latent types can substantially increase output. The study shows reorganizing existing teams leveraging the complementarity between latent types could increase the expected number of successful deals by about 26.6%, whereas reorganizing teams using only observed gender or education information improves performance by just 2.3% and 1.0%, respectively.
- Intermediate solo performers are the best team players, consistently boosting teammates’ performance.
- Top solo performers excel individually but are not the best teammates; in particular, pairing them with bottom solo performers can reduce team output. Pairing two top solo agents also does not generate the highest team output.
Implication: To design effective teams, firms need to look beyond observable demographics and basic performance metrics. Identifying workers’ latent “team types” and the complementarities between them enables managers to reassign existing employees into more productive teams without changing overall headcount.

Source Publication:

Xu, Y., Hu, M., Chu, J., & Ching, A. T. (2025). Heterogeneous complementarity and team design: The case of real estate agents. Marketing Science, 44(3), 626–654.

Background and Research Focus

Effective teamwork is critical in competitive and uncertain markets. Assembling high-performing teams is challenging, however, because traits crucial for collaboration—such as work style and soft skills—are often unobserved by managers. Moreover, firms typically observe only team-level output, making determining which types of workers perform well together even yarder. This study investigates whether unobserved worker types exhibit heterogeneous complementarities and how firms can leverage this hidden variation to systematically improve team design.

Data and Methodology

The analysis draws on proprietary data from Lianjia, a leading Chinese real estate brokerage, covering 484 agents in Beijing’s Beiyuan business zone (2011–2017). The dataset includes complete team-assignment histories, transaction outcomes, and property characteristics. Team assignments were largely exogenous because they depended on worker availability when a property owner arrived and on workload fairness, creating a quasi-random environment suitable for causal inference. The sample includes 52,984 property listings: 77.1% handled by solo performers and 22.9% by two-agent teams.

The authors develop a teamwork model in which agents differ in unobserved characteristics, captured by latent types (with in the main specification). Team production depends on both latent types—which reflect unobserved soft skills—and observed demographics, along with rich property controls. Importantly, the model places no functional-form restrictions on how different types combine within teams; instead, it estimates type and type-pair fixed effects for both solo and two-agent teams, allowing the data to flexibly reveal patterns of heterogeneous complementarity.

Methodologically, the model adapts Bonhomme’s (2021) framework for continuous team-production outcomes to a binary outcome setting, drawing on ideas from stochastic blockmodels for network data. To address the high-dimensional and non-separable likelihood generated by the networked team structure, the authors employ a mean-field variational approximation, enabling feasible estimation while preserving the rich heterogeneity embedded in the team-production process.

Table 1 Agent Types

Type	Solo Performance	Collaborative Ability	Description
1	Very low	Very low	Weak individually and as team players
2	Low	Low to moderate	Gains moderately from strong partners
3	Moderate	Moderate	Performs decently alone and in teams
4	Moderate	Very high	Exceptional collaborators; boost teammates’ outcomes
5	High	Moderate	Strong solo, reasonable team contributor
6	Very high	Low	“Star solo”; excels individually but limited in teams

Findings and Discussion

Although latent types are estimated, agents are labeled as Types 1 through 6—ordered from lowest to highest solo performance—for ease of interpretation.

Intermediate solo performers are the most effective teammates: Type-4 agents, despite only moderate solo performance, generate some of the strongest outcomes in team settings. They form highly productive pairings with many other types, and Type-4–Type-4 teams achieve the highest predicted success rate. This finding indicates strong team players are not necessarily the highest individual performers.

Top solo performers do not translate into top team performers: Type-6 agents excel when working alone, but their advantage diminishes in teams. Their team success rates often fall below their solo performance, and pairing two Type-6 agents is not optimal, suggesting top individual talent may come with weaker collaborative skills.

Matching similar types is not always best: except for the Type-4–Type-4 combination, same-type pairings do not perform best, and pairing types that are too dissimilar can also reduce effectiveness. Teams work best when members differ moderately in latent traits.

Observed demographics matter, but less so: gender mix, age similarity, and education similarity all contribute positively to team outcomes, but these effects are small compared with the complementarities revealed by latent types.

Taken together, the findings underscore that the most effective teams are not composed of the most talented individuals, but of those whose abilities and working styles complement one another.

Implications

The study offers a compelling lesson for organizational design. When success depends on collaboration, firms should look beyond individual performance metrics. Employees with strong collaborative capital—those who consistently enhance their teammates’ productivity—are vital to sustained team success.

Team-formation policies should recognize these less visible but highly valuable contributors. Pairing individuals who balance rather than replicate each other’s strengths can unlock substantial productivity gains. In short, the best teams are not those with the best people, but those with the right mix of people.