I am a problem-driven researcher fascinated by how emerging technologies reshape decisions. My research is guided by questions, and I draw on the analytical and empirical approaches best suited to answering them. My current research spans artificial intelligence, digital manufacturing, electric vehicles, and online platforms, with a common focus on how technological innovations reshape decision making and system performance. Methodologically, I have employed tools including stochastic processes, queueing theory, optimization, stochastic comparison, dynamic programming, game theory, statistical modeling, and field experiment. 

I always look forward to collaborations with industry partners. My contact information can be found on the home page.

Below is a selection of my research work:


Human decision makers (HDMs) increasingly rely on artificial intelligence (AI) systems for decision support in high-stakes contexts, yet how they integrate AI recommendations into their decision making remains insufficiently understood. Observed responses to AI assistance confound how HDMs form beliefs from available information with how they translate those beliefs into decisions, making it difficult to identify the behavioral mechanisms underlying human-AI collaboration. We develop a structural model of AI-augmented human decision making that separates belief formation from choice implementation. The framework combines a general belief-aggregation layer, which allows human and AI evidence to receive different weights and interact, with a stochastic choice layer that captures imperfect translation of beliefs into actions. We estimate the model using experimental data from Reverberi et al. (2022), in which endoscopists classify colonoscopy cases with and without AI assistance. The results show that HDMs incorporate both their own diagnostic evidence and AI recommendations but systematically underweight AI recommendations relative to a normative Bayesian benchmark. Moreover, compared to non-experts, expert HDMs place greater weight on their own diagnostic evidence while exhibiting similar responsiveness to AI recommendations, indicating that i) expertise affects the belief-aggregation process itself, rather than merely the strength of human evidence; ii) expertise appears to moderate AI influence through stronger reliance on private information, rather than through greater algorithm aversion. Finally, by recovering the full choice probability rather than realized decisions only, the model enables evaluation of AI's decision impact beyond observed sample paths and a probabilistic assessment of AI-assisted diagnostic performance.

Gao, Y., Y. Zhang, S. Bansal, D. Guide. 2026. Use-based vs. Time-based: Contract Design and Product Flow Dynamics in OEM Servicizing. Under review. 

To reduce material and energy consumption, many original equipment manufacturers (OEMs) are adopting servicizing operations, whereby products are leased to customers, remanufactured at the end of each lease term, and re-leased to the market. Servicizing contracts are typically structured as either time-based (charging by lease duration) or use-based (charging by product usage). This paper develops a closed queueing network (CQN) model to evaluate how these contract types influence product flow dynamics and cost efficiency. For a homogeneous customer base with a single usage rate, we derive a closed-form distribution of the CQN and show that contract parameters affect costs through the utilization ratio, which governs product flow between leasing and remanufacturing queues. When remanufacturing rates decrease linearly with usage, both contract types perform equivalently; under strictly concave remanufacturing rates, however, use-based contracts are more cost-effective. For heterogeneous customers with two usage rates, the CQN does not admit a closed-form solution. We propose an approximate utilization ratio and demonstrate that use-based contracts remain advantageous even under linear remanufacturing costs, though their relative benefit diminishes with larger remanufacturing capacity or lower customer heterogeneity. Discrete-event simulations validate the approximation and highlight the importance of jointly considering leasing and remanufacturing decisions in servicizing operations. From a managerial perspective, our findings provide actionable guidance for OEMs in designing leasing contracts that balance cost efficiency, customer heterogeneity, and remanufacturing capacity, thereby supporting the broader goals of sustainability and resource efficiency.


To reduce carbon emissions, ride-hailing platforms such as Uber and Lyft have introduced “Green” ride options that allow riders to request environmentally friendly vehicles. We study the economic and environmental implications of such an option on a platform that matches electric vehicle (EV) and fuel vehicle (FV) drivers with heterogeneous riders who differ in their environmental preferences. We develop a stylized model to characterize the platform’s optimal two-sided pricing with and without a “Green” option, subject to an EV penetration target. Without the “Green” option, the platform charges a uniform price to riders but may offer differentiated wages to EV and FV drivers. In this case, even in the absence of binding regulation, riders' environmental preferences induce the platform to attract more EV supply by offering EV drivers higher wages. With the “Green” option, the platform can differentiate prices across green and regular rides. However, separating riders into two matching channels creates a de-pooling effect that reduces matching efficiency. Contrary to conventional wisdom, introducing the “Green” option does not necessarily increase EV penetration and may even worsen the platform's environmental performance. This can occur when eco-conscious riders are relatively scarce or when their environmental preferences are sufficiently strong: the former limits the market for green rides, while the latter allows the platform to charge a substantial green premium that suppresses demand and reduces the need to expand EV supply. A “Green” option improves both platform profit and environmental performance only when eco-conscious riders' environmental preferences are at an intermediate level. Moreover, although more stringent EV penetration targets improve environmental performance, they can reduce both rider and driver welfare, revealing an important policy trade-off between environmental and social objectives.

Zhang, Y., B. Tomlin. 2026. 3D Printer Rentals: Technical Service Design and Pricing for Customers of Varying Expertise. Manufacturing & Service Operations Management. 28(4) 1154-1171


Businesses with short-term production needs can rent equipment (e.g., 3D printer) from a manufacturer or distributor (hereafter, OEM), or use a manufacturing-services firm (e.g., print farm). Those who rent may vary in their expertise and therefore in their cost of attaining quality. Those who use a print farm benefit from the expertise of print-farm technicians. An OEM that offers rentals has an incentive to provide technical service to improve customer expertise so as to make renting attractive. We analytically study the rental-and-technical-service-design problem for an OEM, accounting for the heterogeneity in customer expertise and the choice of utility-maximizing customers to rent or to use a print farm. We consider both a setting where the OEM provides the same menu of offerings to all customers (population-based) and a setting where it customizes the offerings to each customer (user-based). We also study an extension where the print farm actively competes and one where the OEM owns the print farm. We characterize the optimal offering for the OEM, the resulting profit, the value of customization (difference in profits under user- and population-based), and the impact of expertise distribution on these terms. Our problem is related to vertical quality differentiation but leads to several contrasting insights, including that offering a single quality level can be optimal, that the deadweight loss can be lower for a more heterogeneous customer population, and that upward (instead of the classic downward) distortion occurs. We provide guidance on when technical service should be bundled with the rental and when it should be offered as an optional upgrade. Our findings highlight when rental markets with higher heterogeneity are more or less attractive.  Conventional wisdom suggests that higher customer heterogeneity should increase the value of customization. We prove this wisdom is not always correct.