Kenny Ching
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Kenny Ching has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Enrico Forti
Seton Hall University and University College London
2 shared publications Scott Stern
Sloan School of Management, Massachusetts Institute of Technology, MA, USA
2 shared publicationsA. Shaahinfar
1 shared publicationE.D. Gurewitsch
1 shared publicationE.S. Hu
1 shared publication- Evan Rawley
University of Connecticut Stamford Campus, Stamford, Connecticut 06901
1 shared publication
Latest Publications
- Open Access
- 29 June 2026
- Published by Center for Open Science in SocArXiv
Mandatory AI Disclosure Sorts Products into Launch Failure: Evidence from Steam
Platforms increasingly require firms to disclose when products incorporate generative artificial intelligence (AI). Existing evidence largely asks whether such disclosure changes consumer demand; we show that a complementary question—where demand is lost—is at least as important. Using a cross-section of 799 visual-novel games on Steam released early enough to observe a complete 90-day post-release window—of which 22% carry a platform-mandated AI-use disclosure—we show that the consumer response to AI disclosure is concentrated almost entirely on the extensive margin. Disclosed games are roughly twice as likely to obtain no early traction at all: the rate of zero day-30 reviews rises from 5.0% to 9.0%, an odds ratio of 2.14. Among games that do gain an audience, by contrast, the rate of subsequent review accumulation is statistically indistinguishable by disclosure status, and the continuous accumulation penalty is not robust to controls. Disclosed games that are reviewed receive ratings about six percentage points lower. The market response to disclosure is therefore not a uniform discount applied to every product, but a bimodal pattern: a higher probability of failing to launch, and modestly lower satisfaction among survivors. We discuss implications for the design of mandatory disclosure more broadly: when a label can determine whether a product is seen at all, the welfare consequences of disclosure differ sharply depending on whether it gently discounts demand or sorts products into launch failure, and on the producer incentives it sets in motion.
- 29 June 2026
- Published by Center for Open Science in SocArXiv
Open AccessMandatory AI Disclosure Sorts Products into Launch Failure: Evidence from Steam
Platforms increasingly require firms to disclose when products incorporate generative artificial intelligence (AI). Existing evidence largely asks whether such disclosure changes consumer demand; we show that a complementary question—where demand is lost—is at least as important. Using a cross-section of 799 visual-novel games on Steam released early enough to observe a complete 90-day post-release window—of which 22% carry a platform-mandated AI-use disclosure—we show that the consumer response to AI disclosure is concentrated almost entirely on the extensive margin. Disclosed games are roughly twice as likely to obtain no early traction at all: the rate of zero day-30 reviews rises from 5.0% to 9.0%, an odds ratio of 2.14. Among games that do gain an audience, by contrast, the rate of subsequent review accumulation is statistically indistinguishable by disclosure status, and the continuous accumulation penalty is not robust to controls. Disclosed games that are reviewed receive ratings about six percentage points lower. The market response to disclosure is therefore not a uniform discount applied to every product, but a bimodal pattern: a higher probability of failing to launch, and modestly lower satisfaction among survivors. We discuss implications for the design of mandatory disclosure more broadly: when a label can determine whether a product is seen at all, the welfare consequences of disclosure differ sharply depending on whether it gently discounts demand or sorts products into launch failure, and on the producer incentives it sets in motion.
- Open Access
- 22 June 2026
- Published by Center for Open Science in SocArXiv
Challengers Shift to Lower-Tier Competition When Elite Incumbents Extend Their Tenure: Evidence From Professional Tennis
Extended incumbent tenure is hard to study empirically because dominant incumbents and weaker successors may simply differ in talent. We study this using professional tennis, where theBig Four—Federer, Nadal, Djokovic, and Murray—remained at the summit 5–10 years beyond the historical norm. For 296 players with fully elapsed prime windows, we construct a tenureextension exposure index measuring how far past historical norms the Big Four were during each player’s prime. Controlling for pre-prime talent, higher exposure predicts fewer main-tour matches, more Challenger-circuit matches, and earlier exit from the world top 100 once secular trends are absorbed—consistent with downward substitution from elite to sub-elite competition. Raw Big Four dominance shows the opposite pattern, indicating the effect is specific to extended tenure, not opponent quality. Talented athletes are not simply outcompeted but redirected to lower-visibility, lower-revenue tiers, with implications for tournament design, athlete welfare, and succession planning where senior careers are lengthening.
- 22 June 2026
- Published by Center for Open Science in SocArXiv
Open AccessChallengers Shift to Lower-Tier Competition When Elite Incumbents Extend Their Tenure: Evidence From Professional Tennis
Extended incumbent tenure is hard to study empirically because dominant incumbents and weaker successors may simply differ in talent. We study this using professional tennis, where theBig Four—Federer, Nadal, Djokovic, and Murray—remained at the summit 5–10 years beyond the historical norm. For 296 players with fully elapsed prime windows, we construct a tenureextension exposure index measuring how far past historical norms the Big Four were during each player’s prime. Controlling for pre-prime talent, higher exposure predicts fewer main-tour matches, more Challenger-circuit matches, and earlier exit from the world top 100 once secular trends are absorbed—consistent with downward substitution from elite to sub-elite competition. Raw Big Four dominance shows the opposite pattern, indicating the effect is specific to extended tenure, not opponent quality. Talented athletes are not simply outcompeted but redirected to lower-visibility, lower-revenue tiers, with implications for tournament design, athlete welfare, and succession planning where senior careers are lengthening.
- Open Access
- 13 May 2026
- Published by Center for Open Science in SocArXiv
The Epistemic Divorce: Empirical Evidence for the Systematic Decoupling of Academic Research from Industrial Innovation
We document the systematic decoupling of academic research from industrial innovation across four major scientific fields using 1,976,455 US patents linked to academic papers through applicant-added citations. Measuring industry reliance as patent citations per academic paper across 20 subfields in computer science, biology, medicine, and materials science from 1980 to 2019, we find a universal inverted-U pattern: industry reliance rises during an initial period of joint exploration, peaks, then declines by 4795% and does not recover. Structural break dates are staggered within each field in theoretically predicted order of commercial maturity, and across fields in a consistent sequence: computer science and medicine divorce earliest (mean peak ≈1993), followed by biology (≈1998) and materials science (≈1997). Two subfields at the analysis window boundary (Virology, Chemical Engineering) are flagged as potentially unreliable and excluded from cross-field inference. The post-2012 deep learning revolution did not reverse the divorce in any CS subfield, suggesting that technological revolutions do not automatically re-couple academia and industry once the frontier has migrated inside proprietary firms. These findings provide the first systematic cross-field empirical confirmation of a predictable knowledge appropriability cycle, with substantial implications for science policy and public R&D investment.
- 13 May 2026
- Published by Center for Open Science in SocArXiv
Open AccessThe Epistemic Divorce: Empirical Evidence for the Systematic Decoupling of Academic Research from Industrial Innovation
We document the systematic decoupling of academic research from industrial innovation across four major scientific fields using 1,976,455 US patents linked to academic papers through applicant-added citations. Measuring industry reliance as patent citations per academic paper across 20 subfields in computer science, biology, medicine, and materials science from 1980 to 2019, we find a universal inverted-U pattern: industry reliance rises during an initial period of joint exploration, peaks, then declines by 4795% and does not recover. Structural break dates are staggered within each field in theoretically predicted order of commercial maturity, and across fields in a consistent sequence: computer science and medicine divorce earliest (mean peak ≈1993), followed by biology (≈1998) and materials science (≈1997). Two subfields at the analysis window boundary (Virology, Chemical Engineering) are flagged as potentially unreliable and excluded from cross-field inference. The post-2012 deep learning revolution did not reverse the divorce in any CS subfield, suggesting that technological revolutions do not automatically re-couple academia and industry once the frontier has migrated inside proprietary firms. These findings provide the first systematic cross-field empirical confirmation of a predictable knowledge appropriability cycle, with substantial implications for science policy and public R&D investment.
- Open Access
- 12 May 2026
- Published by Springer Nature in Research Square
The Epistemic Divorce: Empirical Evidence for the Systematic Decoupling of Academic Research from Industrial Innovation
We document the systematic decoupling of academic research from industrial innovation across four major scientic fields using 1,976,455 US patents linked to academic papers through applicant-added citations. Measuring industry reliance as patent citations per academic paper across 20 subfields in computer science, biology, medicine, and materials science from 1980 to 2019, we find a universal inverted-U pattern: industry reliance rises during an initial period of joint exploration, peaks, then declines by 4795% and does not recover. Structural break dates are staggered within each field in theoretically predicted order of commercial maturity, and across fields in a consistent sequence: computer science and medicine divorce earliest (mean peak ≈1993), followed by biology (≈1998) and materials science (≈1997). Two subelds at the analysis window boundary (Virology, Chemical Engineering) are agged as potentially unreliable and excluded from cross-eld inference. The post-2012 deep learning revolution did not reverse the divorce in any CS subfield, suggesting that technological revolutions do not automatically re-couple academia and industry once the frontier has migrated inside proprietary firms. These findings provide the first systematic cross-field empirical confirmation of a predictable knowledge appropriability cycle, with substantial implications for science policy and public R&D investment.
- 12 May 2026
- Published by Springer Nature in Research Square
Open AccessThe Epistemic Divorce: Empirical Evidence for the Systematic Decoupling of Academic Research from Industrial Innovation
We document the systematic decoupling of academic research from industrial innovation across four major scientic fields using 1,976,455 US patents linked to academic papers through applicant-added citations. Measuring industry reliance as patent citations per academic paper across 20 subfields in computer science, biology, medicine, and materials science from 1980 to 2019, we find a universal inverted-U pattern: industry reliance rises during an initial period of joint exploration, peaks, then declines by 4795% and does not recover. Structural break dates are staggered within each field in theoretically predicted order of commercial maturity, and across fields in a consistent sequence: computer science and medicine divorce earliest (mean peak ≈1993), followed by biology (≈1998) and materials science (≈1997). Two subelds at the analysis window boundary (Virology, Chemical Engineering) are agged as potentially unreliable and excluded from cross-eld inference. The post-2012 deep learning revolution did not reverse the divorce in any CS subfield, suggesting that technological revolutions do not automatically re-couple academia and industry once the frontier has migrated inside proprietary firms. These findings provide the first systematic cross-field empirical confirmation of a predictable knowledge appropriability cycle, with substantial implications for science policy and public R&D investment.
- Open Access
- 23 April 2026
- Published by Center for Open Science in SocArXiv
A Market for Preprints: Knowledge Accessibility, Execution Leakage, and the Economics of Academic Disclosure in the Age of AI
Research fields differ in knowledge accessibility — the degree to which a field’s questions and execution methods are legible to trained non-specialists. We develop a sequential game-theoretic model of preprint disclosure in which researcher A chooses whether to declare an idea; if she declares, researcher B learns the idea and executes with probability λ, the execution leakage rate. The declarer retains residual priority share α; the rival incurs execution cost ce. A single cutoff λc separates universal disclosure from strategic opacity. Generative AI raises λ heterogeneously — sharply in high-accessibility fields where execution requires only codified knowledge, negligibly where execution requires tacit expertise or physical capital AI cannot substitute. Heterogeneousidea value generates adverse selection: as λ rises, the disclosure cutoff π∗ rises, withdrawing high-value ideas while AI-lowered production costs sustain a flood of lower-value declarations — the compositional flood. Using OpenAlex data across seven fields from 2019 to 2024, we document a structural break: low-accessibility fields (mathematics, physics, chemistry) decelerate post-ChatGPT (mean −1.22 pp/year) while high-accessibility fields (economics, sociology) accelerate sharply (+1.51 pp/year) — a differential of 2.73 pp/year coinciding precisely with the AI transition. We identify a new pre-publication anti-commons in the tradition of Murray & Stern (2007) and discuss implications for preprint platform design and open science policy.
- 23 April 2026
- Published by Center for Open Science in SocArXiv
Open AccessA Market for Preprints: Knowledge Accessibility, Execution Leakage, and the Economics of Academic Disclosure in the Age of AI
Research fields differ in knowledge accessibility — the degree to which a field’s questions and execution methods are legible to trained non-specialists. We develop a sequential game-theoretic model of preprint disclosure in which researcher A chooses whether to declare an idea; if she declares, researcher B learns the idea and executes with probability λ, the execution leakage rate. The declarer retains residual priority share α; the rival incurs execution cost ce. A single cutoff λc separates universal disclosure from strategic opacity. Generative AI raises λ heterogeneously — sharply in high-accessibility fields where execution requires only codified knowledge, negligibly where execution requires tacit expertise or physical capital AI cannot substitute. Heterogeneousidea value generates adverse selection: as λ rises, the disclosure cutoff π∗ rises, withdrawing high-value ideas while AI-lowered production costs sustain a flood of lower-value declarations — the compositional flood. Using OpenAlex data across seven fields from 2019 to 2024, we document a structural break: low-accessibility fields (mathematics, physics, chemistry) decelerate post-ChatGPT (mean −1.22 pp/year) while high-accessibility fields (economics, sociology) accelerate sharply (+1.51 pp/year) — a differential of 2.73 pp/year coinciding precisely with the AI transition. We identify a new pre-publication anti-commons in the tradition of Murray & Stern (2007) and discuss implications for preprint platform design and open science policy.
- Open Access
- 8 April 2026
- Published by Center for Open Science in SocArXiv
When Complementary Assets Are Free: GPT Platforms and the Transformation of Startup Commercialization Strategy
The seminal framework of Gans & Stern (2003) identifies the ownership of specialized complementary assets by incumbents as the central driver of startup commercialization strategy, predicting cooperation between innovators and established firms when such assets are critical. This paper extends the framework to account for a structural feature of the contemporary innovation landscape that the original analysis did not contemplate: the emergence of General Purpose Technology (GPT) platforms that provide high-quality complementary assets freely, non-exclusively, and with automatic improvement over time. We develop an analytical framework in which a GPT platform constitutes a third source of complementary assets --- distinct from both the incumbent firm and independent commercialization --- and derive conditions under which this transforms the cooperation--competition tradeoff at the heart of the Gans--Stern framework. Four results emerge. First, above a platform quality threshold, the cooperation surplus collapses and no mutually beneficial agreement between innovator and incumbent is feasible, even when incumbents hold substantial complementary asset stocks. Second, incumbents with larger complementary asset stocks are predicted to perform worse in the GPT platform regime, inverting the standard prediction. Third, a distinct commercialization mode --- platform riding --- emerges as the dominant strategy in a well-defined parameter region, constituting a necessary extension to the Gans--Stern two-by-two commercialization matrix. Fourth, GPT platform emergence strictly improves welfare for the set of innovations previously suppressed by incumbent hold-up, with gains largest where the cooperation problem was most acute. We illustrate the framework using matched incumbent--entrant pairs in the Internet of Things sector, including the contrast between SwingVision and Babolat in smart tennis equipment, and discuss implications for innovation policy in platform-mediated economies.
- 8 April 2026
- Published by Center for Open Science in SocArXiv
Open AccessWhen Complementary Assets Are Free: GPT Platforms and the Transformation of Startup Commercialization Strategy
The seminal framework of Gans & Stern (2003) identifies the ownership of specialized complementary assets by incumbents as the central driver of startup commercialization strategy, predicting cooperation between innovators and established firms when such assets are critical. This paper extends the framework to account for a structural feature of the contemporary innovation landscape that the original analysis did not contemplate: the emergence of General Purpose Technology (GPT) platforms that provide high-quality complementary assets freely, non-exclusively, and with automatic improvement over time. We develop an analytical framework in which a GPT platform constitutes a third source of complementary assets --- distinct from both the incumbent firm and independent commercialization --- and derive conditions under which this transforms the cooperation--competition tradeoff at the heart of the Gans--Stern framework. Four results emerge. First, above a platform quality threshold, the cooperation surplus collapses and no mutually beneficial agreement between innovator and incumbent is feasible, even when incumbents hold substantial complementary asset stocks. Second, incumbents with larger complementary asset stocks are predicted to perform worse in the GPT platform regime, inverting the standard prediction. Third, a distinct commercialization mode --- platform riding --- emerges as the dominant strategy in a well-defined parameter region, constituting a necessary extension to the Gans--Stern two-by-two commercialization matrix. Fourth, GPT platform emergence strictly improves welfare for the set of innovations previously suppressed by incumbent hold-up, with gains largest where the cooperation problem was most acute. We illustrate the framework using matched incumbent--entrant pairs in the Internet of Things sector, including the contrast between SwingVision and Babolat in smart tennis equipment, and discuss implications for innovation policy in platform-mediated economies.
- Open Access
- 25 March 2026
- Published by Center for Open Science in MediArXiv
Moral Memory Entrepreneurship: Collective Remembering and Forgetting in the Digital Creator Economy
Digital entrepreneurship increasingly unfolds in spaces with little formal regulation, raising new ethical questions about how norms of responsibility emerge and endure. This paper examines how a digital creator community reconstructed its moral order following controversy. Using a longitudinal digital ethnography of ninety public artefacts, the study identifies a four-phase process—exposure, amplification, repair, and reconstruction—through which a collective re-made its moral past. We conceptualise this as moral memory entrepreneurship, the entrepreneurial reconstruction of collective memory to define acceptable conduct in post-institutional contexts. The findings extend theories of moral entrepreneurship and collective memory by showing how rivals, audiences, and algorithms serve as mnemonic agents. The paper concludes by proposing the principle of responsible remembering—balancing the moral duties to remember and to forget—as central to ethical governance in the digital age.
- 25 March 2026
- Published by Center for Open Science in MediArXiv
Open AccessMoral Memory Entrepreneurship: Collective Remembering and Forgetting in the Digital Creator Economy
Digital entrepreneurship increasingly unfolds in spaces with little formal regulation, raising new ethical questions about how norms of responsibility emerge and endure. This paper examines how a digital creator community reconstructed its moral order following controversy. Using a longitudinal digital ethnography of ninety public artefacts, the study identifies a four-phase process—exposure, amplification, repair, and reconstruction—through which a collective re-made its moral past. We conceptualise this as moral memory entrepreneurship, the entrepreneurial reconstruction of collective memory to define acceptable conduct in post-institutional contexts. The findings extend theories of moral entrepreneurship and collective memory by showing how rivals, audiences, and algorithms serve as mnemonic agents. The paper concludes by proposing the principle of responsible remembering—balancing the moral duties to remember and to forget—as central to ethical governance in the digital age.
- Open Access
- 24 March 2026
- Published by Springer Nature in Research Square
Behavioural Transparency in Multi-Agent RL: Strategic Decoupling and the Economic Optimisation of Tactical Shocks
A primary challenge in deploying multi-agent reinforcement learning (RL) systems is the opacity of their emergent strategies, necessitating frameworks that render complex agentic behaviour interpretable. When RL agents interact with high-dimensional environments, they often execute decisions that appear erratic or "alien" to human observers. Behavioural economic theory suggests this perception stems from tactical myopia — the tendency of bounded biological agents to treat localised shocks (such as a tactical loss or victory) as terminal states, thereby degrading subsequent macroeconomic efficiency. Utilising high-fidelity telemetry from OpenAI Five in the imperfect-information environment of Dota 2, we provide empirical transparency into the processing of these tactical shocks. We demonstrate that average human cohorts exhibit intense tactical myopia: localised failures trigger extended economic contraction, while localised successes induce satisficing and stagnant resource acquisition. Conversely, econometric state-space matching reveals that the RL agent executes perfect strategic decoupling, treating shocks as neutral state transitions to instantly optimise subsequent macroeconomic trajectories. Crucially, when the RL agent is compared exclusively against apex human experts in identical economic states, the statistical divergence between biological and synthetic agents largely vanishes at all economically meaningful horizons. This convergence provides a transparent econometric explanation for opaque AI behaviour: RL networks do not invent incomprehensible strategies; rather, they mathematically purge the tactical myopia inherent in average play, converging on the precise global optimisation phenotype utilised by apex biological experts.
- 24 March 2026
- Published by Springer Nature in Research Square
Open AccessBehavioural Transparency in Multi-Agent RL: Strategic Decoupling and the Economic Optimisation of Tactical Shocks
A primary challenge in deploying multi-agent reinforcement learning (RL) systems is the opacity of their emergent strategies, necessitating frameworks that render complex agentic behaviour interpretable. When RL agents interact with high-dimensional environments, they often execute decisions that appear erratic or "alien" to human observers. Behavioural economic theory suggests this perception stems from tactical myopia — the tendency of bounded biological agents to treat localised shocks (such as a tactical loss or victory) as terminal states, thereby degrading subsequent macroeconomic efficiency. Utilising high-fidelity telemetry from OpenAI Five in the imperfect-information environment of Dota 2, we provide empirical transparency into the processing of these tactical shocks. We demonstrate that average human cohorts exhibit intense tactical myopia: localised failures trigger extended economic contraction, while localised successes induce satisficing and stagnant resource acquisition. Conversely, econometric state-space matching reveals that the RL agent executes perfect strategic decoupling, treating shocks as neutral state transitions to instantly optimise subsequent macroeconomic trajectories. Crucially, when the RL agent is compared exclusively against apex human experts in identical economic states, the statistical divergence between biological and synthetic agents largely vanishes at all economically meaningful horizons. This convergence provides a transparent econometric explanation for opaque AI behaviour: RL networks do not invent incomprehensible strategies; rather, they mathematically purge the tactical myopia inherent in average play, converging on the precise global optimisation phenotype utilised by apex biological experts.
- 1 July 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Organization Science
- Vol. 35 (4) , 1409-1426
Competitive Familiarity: Learning to Coordinate by Competing
This paper develops and tests a theory of organizational learning, proposing that prior competitive interaction improves coordination among teammates. We test the theory using millions of experiments in the formation of eSports teams. The results show that exogenously assigned teams of former competitors are highly effective—The marginal returns to prior competitive interaction are even larger than the returns to prior collaborative interaction. The evidence suggests that teammates learn to coordinate by competing, a finding with implications for organizational design and the management of human capital. Funding: The authors are grateful to UCL School of Management and Worcester Polytechnic Institute for financial and technical support.
- 1 July 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Organization Science
- Vol. 35 (4) , 1409-1426
Competitive Familiarity: Learning to Coordinate by Competing
This paper develops and tests a theory of organizational learning, proposing that prior competitive interaction improves coordination among teammates. We test the theory using millions of experiments in the formation of eSports teams. The results show that exogenously assigned teams of former competitors are highly effective—The marginal returns to prior competitive interaction are even larger than the returns to prior collaborative interaction. The evidence suggests that teammates learn to coordinate by competing, a finding with implications for organizational design and the management of human capital. Funding: The authors are grateful to UCL School of Management and Worcester Polytechnic Institute for financial and technical support.
- 1 January 2021
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Organization Science
- Vol. 32 (1) , 1-17
Extemporaneous Coordination in Specialist Teams: The Familiarity Complementarity
Team production is ubiquitous in the economy, but managing teams effectively remains a challenge for many organizations. This paper studies how familiarity among teammates influences the performance of specialist teams, relative to nonspecialist teams. Applying theories of team production to contexts where team members coordinate interdependent activities extemporaneously, we develop predictions about factors that shift the marginal returns to specialization along two dimensions of familiarity: social familiarity and functional familiarity. We test our hypotheses in the context of Defence of the Ancients 2 (DOTA2), a major e-sports game where, in some formats, players are exogenously assigned to five-person teams. After analyzing nearly 6.5 million matches, we find that specialist teams are relatively more successful when members are more socially and functionally familiar with one another. The results suggest that the plug-and-play perspective on specialist teams is incomplete; rather, specialization and familiarity are complements in dynamic environments where team members coordinate extemporaneously. Funded: Financial support from the UCL School of Management and Worcester Polytechnic Institute is gratefully acknowledged.
- 1 January 2021
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Organization Science
- Vol. 32 (1) , 1-17
Extemporaneous Coordination in Specialist Teams: The Familiarity Complementarity
Team production is ubiquitous in the economy, but managing teams effectively remains a challenge for many organizations. This paper studies how familiarity among teammates influences the performance of specialist teams, relative to nonspecialist teams. Applying theories of team production to contexts where team members coordinate interdependent activities extemporaneously, we develop predictions about factors that shift the marginal returns to specialization along two dimensions of familiarity: social familiarity and functional familiarity. We test our hypotheses in the context of Defence of the Ancients 2 (DOTA2), a major e-sports game where, in some formats, players are exogenously assigned to five-person teams. After analyzing nearly 6.5 million matches, we find that specialist teams are relatively more successful when members are more socially and functionally familiar with one another. The results suggest that the plug-and-play perspective on specialist teams is incomplete; rather, specialization and familiarity are complements in dynamic environments where team members coordinate extemporaneously. Funded: Financial support from the UCL School of Management and Worcester Polytechnic Institute is gratefully acknowledged.