Working Papers
Last-Step Human Involvement: The Impact of Industrial GPT on B2B Procurement – Evidence from a Field Experiment
with Shichen Zhang, Xiande Zhao, and Yinliang (Ricky) Tan
​This study examines the impact of Industrial AI agents on Business-to-Business (B2B) procurement, with a focus on the Maintenance, Repair, and Operations (MRO) sector. MRO procurement, characterized by product complexity and specification challenges, remains a critical yet inefficient area of B2B operations. Using large language models, ZKH—a leading digital platform for industrial supplies—has developed an AI-powered agent that acts as an intelligent assistant. By analyzing a database of over 17 million SKUs and billions of product parameters, the agent provides smart recommendations, real-time specification support, and streamlined navigation through complex product catalogs. We conduct a field experiment to evaluate the agent's effects on early-stage buyer engagement and final-stage purchasing behavior. Preliminary results from 254 purchasing companies over a three-month period indicate that, while the agent significantly improves engagement during the initial stages, such as browsing and searching, it does not directly increase online purchases. Instead, it drives greater reliance on human-assisted offline channels for final purchase decisions, emphasizing the critical role of human interaction in the complex B2B transactions. In summary, the agent can facilitate early-stage engagement and enhance product diversity; however, final purchasing decisions still require human involvement. This study underscores the complementary roles of AI and human expertise in B2B procurement, advocating for a hybrid approach that integrates AI-driven efficiency with the relationship-building and problem-solving capabilities of human professionals. The findings offer valuable insights for developing adaptive, buyer-centered procurement strategies to address persistent inefficiencies in the industrial sector.​
Which Jobs Scale and Why?
with Erik Brynjolfsson and Seth Benzell
While some jobs, like CEOs', involve tasks that can change the productivity of the entire organization, others, like piece-rate manufacturing workers, do not. In this paper, we measure which jobs scale and why using a sample of large-scale administrative data from the payroll processing company ADP, covering between 14.7 and 16.0 million employees in each year at 168,371 firms across 308,960 establishments from 2017 to 2023. The median worker in a occupation-firm-year is paid 8.6\% more at an firm with a payroll twice as large. To evaluate different theories of this premium, we investigate heterogeneity in wage scaling by job characteristics. We find wages scale the most with firm size for workers in managerial, decision making, and abstract tasks, and less for workers in routine and manual tasks. Wage scaling is also stronger for workers who are at high percentiles of wage in their occupation-organization-year and in IT intensive firms. IT intensity at firms is related to larger wage scaling for abstract and top percentile workers, and smaller scaling for routine, manual and low-percentile workers, effects which intensify the direct effect of scaling for these jobs. Taken together, these results are most consistent with `Span-of-Control'-driven wage scaling. This implies that the disproportionate growth of management-intensive occupations and the rise of large and IT intensive firms will tend to increase aggregate wage polarization.
​
Invited presentations:
-
Workshop on Information Systems and Economics (WISE) (India, Dec 2023)
-
NBER Mega Firm and the Economy Conference (scheduled September 2024, Cambridge, MA)
​
IT and Innovation: How did the Internet affect firms’ reliance on science?
This paper examines how the Internet facilitates the utilization of science in industrial innovation. I find that the Internet enables firms to discover “hidden gems”- commercializable yet under-recognized scientific findings in less prestigious journals, from early-career scientists, with fewer academic citations, and with higher forward patent citations. I compiled a database that contains 541,568 patent citations to scientific papers from 3,651 public firm-locations between 1992 and 2000, and identified the staggered adoption of basic Internet at these firms. Using a difference-in-differences framework, I show that access to the Internet at firm-locations is associated with a 9.3% increase in the likelihood of citing scientific papers; and up to 13.2% increase in citing the "hidden gem" papers. These findings suggest that IT reshapes the process of firm sourcing knowledge in innovation. By reducing search cost, IT enables firms to have equal access to previously less noticeable scientific knowledge and thus discover their commercial value. The results shed light on how IT reinforces the link between science and innovation.
​
Invited presentations:
-
Cornell Innovation, Entrepreneurship, and Technology Brownbag Workshop (Ithaca, NY, Sep 2021)
-
Cornell Strategy and Business Economics Workshop (Ithaca, Oct 2021)
-
Information Systems Student Presentations Over the Cloud Workshop (ISPOC) (Virtual, Nov 2021)
-
the 14th Workshop on the "Organisation, Economics and Policy of Scientific Research" WOEPSR22 (Leuven, Belgium, April 2022)
-
the 2nd annual International Conference on the Science of Science and Innovation (ICSSI) (Northwestern University, Evanston, IL, June 2023)
How does labor mobility affect business adoption of a new technology? The case of machine learning
with Natarajan Balasubramanian and Chris Forman, Strategic Management Journal (2024)
Full paper​ Online appendix
We investigate how worker mobility influences the adoption of a new general-purpose technology (GPT). Using data from over 153,000 establishments between 2010 and 2018, we observe establishment decisions to adopt machine learning. Taking advantage of state-level changes to the enforceability of noncompete agreements as an exogenous shock to worker mobility, we find that changes that facilitate worker movements are associated with a significant decline in the likelihood of adoption. Moreover, the magnitude of establishment response depends upon characteristics of the establishment and the location in which it resides, in particular, establishment size and number of large establishments in the same industry-location. These results are consistent with the view that increases in worker mobility lead to greater risks for establishments that are contemplating adoption of a new GPT that involves significant downstream innovation.
​
Invited presentations:
Workshop on Information Systems and Economics (WISE) (Virtual, Dec 2020)
Temple-CMU-NYU 2020 Conference on Artificial Intelligence, Machine Learning, and Business
Analytics (Virtual, Dec 2020)
ISB 2nd AI & Strategy Consortium (Virtual, Jan 2021)
Wharton Innovation Doctoral Symposium (WINDS) (Virtual, Feb 2021)
the 19th ZEW Conference on the Economics of Information and Communication Technologies (Virtual, June 2021)
AOM Symposium- Machine Learning, Artificial Intelligence, and Strategy: Emerging Research on
the Importance of Complements (Virtual, Aug 2021)
2021 NBER Economics of Artificial Intelligence Conference (Virtual, Sep 2021)
Machine Learning, Artificial Intelligence, and Strategy: Research on the Importance of Complements
with Natarajan Balasubramanian, Christopher Forman, Aija Elina Leiponen, Prithwiraj Choudhury, Kristina Steffenson McElheran, Robert Channing Seamans, Ryan Allen, Stephen Michael Impink and Wang Jin. Proceedings of the Academy of Management (2021)
Artificial intelligence (AI) and machine learning (ML) represent general purpose technologies that are rapidly diffusing among businesses. These technologies have the potential to transform industries and to impact the performance of firms. They also present important challenges for managers. Firms investing in general purpose technologies like AI require complements to realize value from them and to align with unique firm needs. In this symposium we bring together four papers that examine various aspects of the diffusion of impact of AI and ML in businesses and how these are affected by the presence of complements at the individual, organizational, and ecosystem level. Together these papers will shed new light on the implications of managerial decisions related to this important set of technologies.
The impact of high-speed railway (HSR) expansion on entrepreneurial firm dynamics: Evidence from China
This paper investigates the impact of China’s high-speed railway (HSR) expansion on its entrepreneurial activities using firm registration data between 2011 to 2015. I find that connecting to HSR benefits mega-cities, while has lead a reduction in new firm entry in smaller cities. To address the non-random railway station placement problem, I constructed an instrumental variable of a hypothetical HSR station network that subjects to global construction cost minimization. I also adopted a market access approach similar to Donaldson (2018), where I calculated the impact of HSR on each city by capturing the changes in all its market access using a reduced-form expression derived from general equilibrium trade theory. I demonstrate that non-connection-induced market access significantly increases firm entry in mega cities by 2%, while decreases firm entry in Tier-3 and Tier-4 cities by up to 5%.