Forging the Future of Work
Backfiring AI? AI Deployment in Workplace
Management Science
AI in the workplace has the potential to change the competitive dynamics among employees. The AI system can learn from high-performing employees and make that knowledge available to others. In a competitive environment, this can disincentivize high-performing employees and ultimately backfire, leading to a decline in overall firm productivity. Our results suggest that some ostensibly simple solutions, such as guaranteeing or increasing wages for adversely affected employees, may not effectively solve the problem, and firms would have to judiciously choose optimal AI efficacy levels to achieve better outcomes.
Di Yuan (Assistant Professor, Auburn University), Manmohan Aseri (Assistant professor, University of Maryland), Narayan Ramasubbu (Professor, University of Pittsburgh)
Learning from Earnings Calls: Graph-Based Conversational Modeling for Financial Prediction
Information Systems Research
Earnings conference calls are valuable venues for business communication. Empirical research has shown that the content of earnings calls contains predictive signals about future market risks, which has motivated a line of computational studies that utilize earnings transcripts for financial forecasting tasks. However, earnings call transcripts are typically very long, and the predictive signals within them are often sparsely distributed across different sections of the transcript. As a result, existing computational methods often fail to capture the essential information within the transcript that is relevant to market risks. In this work, we design a novel method to model earnings transcripts as a conversational graph where graph nodes represent discussed topics and graph edges indicate the similarity between topics. By doing so, we aim to explicitly model what is discussed (i.e., topical content), how it is discussed (e.g., cross-referencing or newly introduced topics), and in what manner it is discussed (e.g., sentiment and other linguistic features) within the transcript. We then leverage a graph neural network to learn transcript-level representations for financial risk forecasting. Experimental results show that the proposed method significantly reduces risk forecasting errors, demonstrating its capability of modeling earnings call transcripts. Moreover, this predictive power holds even after considering the firm’s fundamentals, and the performance gain over baseline models continues to grow as transcript length increases. The interpretability analysis shows that the proposed method identifies cross-referencing and newly introduced topics as influential for risk prediction. Moreover, the model tends to associate transcripts with a higher number of new topics in the Q&A section, more cross-referencing discussions, and more positive sentiment with lower predicted financial risks. This work contributes methodologically by proposing a novel predictive approach specifically tailored to the domain-specific challenge of transcript-based risk forecasting. We also discuss key design insights and implications.
Yi Yang (HKUST), Yixuan Tang (HKUST), Yangyang Fan (HK PolyU), and Kunpeng Zhang (UMD)
The GenAI Future of Consumer Research
Journal of Consumer Research
We develop a novel generative AI (GenAI) trajectory, “democratization-average trap-model collapse,” to identify data and model challenges posed by GenAI, from which we project the GenAI future of consumer research. This trajectory consists of three key phenomena: democratization broadens consumer participation, the average trap produces generic responses, and model collapses occurs when GenAI outputs lose human sensibilities. Data and model challenges arise as democratization enhances data representation but embeds real-world biases; the average trap, caused by next-token prediction algorithms, leads to generic outputs that lack individuality; and model collapse occurs when GenAI increasingly learns from its own outputs, amplifying machine bias and diverging from human behavior. To address these challenges, researchers can leverage democratization to study marginalized consumers and prioritize human-centered research over purely data-driven methods. The average trap can be mitigated by fine-tuning models with task-specific and marginalized consumption data while engineering responses for uniqueness. Preventing model collapse requires integrating human-machine hybrid data and applying theories of mind to realign AI with human-centric consumption. Finally, we outline three future research directions: preserving data distribution tails to support consumption democratization, countering the average trap in next-token prediction, and reversing the trajectory from democratization to model collapse.
Ming-Hui Huang, Distinguished University Professor, National Taiwan University
Roland T. Rust, Distinguished Professor and David Bruce Smith Chair in Marketing, University of Maryland
Prompt Adaptation as a Dynamic Complement in Generative AI Systems
As generative AI systems rapidly improve, a key question emerges: How do users keep up—and what happens if they fail to do so. Drawing on theories of dynamic capabilities and IT complements, we examine prompt adaptation—the adjustments users make to their inputs in response to evolving model behavior—as a mechanism that helps determine whether technical advances translate into realized economic value. In a preregistered online experiment with 1,893 participants, who submitted over 18,000 prompts and generated more than 300,000 images, users attempted to replicate a target image in 10 tries using one of three randomly assigned models: DALL-E 2, DALL-E 3, or DALL-E 3 with automated prompt rewriting. We find that users
with access to DALL-E 3 achieved higher image similarity than those with DALL-E 2—but only about half of this gain (51%) came from the model itself. The other half (49%) resulted from users adapting their prompts in response to the model’s capabilities. This adaptation emerged across the skill distribution, was driven by trial-and-error, and could not be replicated by automated prompt rewriting, which erased 58% of the performance improvement associated with DALL-E 3. Our findings position prompt adaptation as a dynamic complement to generative AI—and suggest that without it, a substantial share of the economic value created when models advance may go unrealized.
Eaman Jahani, UMD
Benjamin Manning, MIT
Hong-Yi TuYe, MIT
Mohammed Alsobay, MIT
Christos Nicolaides, University of Cyprus
Siddharth Suri, Microsoft Research
David Holtz, Columbia
Biodiversity Entrepreneurship
Review of Finance
We study an emerging class of start-up organizations focused on biodiversity conservation and the challenges they face in financing these ventures. Using a novel machine learning method, we identify 630 biodiversity-linked start-ups in PitchBook and compare their financing dynamics to other ventures. Biodiversity start-ups raise less capital but attract a broader coalition of investors, including not only venture capitalists (“value investors”) but also mission-aligned impact funds and public institutions (“values investors”). Values investors provide incremental capital rather than substituting value investors, but funding gaps persist. We show biodiversity-linked start-ups use social media activity to help connect with value investors. Our findings can inform policy and practice for mobilizing private capital toward biodiversity preservation, emphasizing hybrid financing models and strategic communication.
Sean Cao, Robert H. Smith School of Business, University of Maryland
Analytics for Finance and Accounting: Data Structures and Applied AI
Analytics for Finance and Accounting: Data Structures and Applied AI bridges the gap between technical data science education and domain-specific applications in accounting and finance. Designed for students and instructors seeking practical exposure to AI-driven financial analytics, the book prioritizes understanding real-world business data—structured and unstructured—before introducing machine learning techniques. It empowers learners to apply AI tools, such as GPT and pre-trained language models, to analyze corporate disclosures, earnings calls, ESG reports, and other financial documents. Minimizing programming prerequisites, the book integrates video tutorials and applied projects to support hands-on learning. It serves as both a foundational text for graduate-level data analytics courses and a modular supplement for traditional finance and accounting curricula. By combining domain expertise with modern computational tools, this book equips the next generation of financial professionals with the skills to thrive in a data-intensive economy.
Sean Cao, Associate Professor, Robert H. Smith School of Business, University of Maryland, United States of America
AI for Customer Journeys: A Transformer Approach
Journal of Marketing Research
AI for Customer Journeys: A Transformer Approach
Zipei Lu and P. K. Kannan, Smith School of Business, University of Maryland
(forthcoming Journal of Marketing Research)
AI for Customer Journeys: A Transformer Approach introduces a novel artificial intelligence (AI) framework for modeling customer journeys in digital marketing. Leveraging transformer-based models – originally developed for natural language processing - this approach analyzes complex sequences of customer interactions across multiple channels (e.g., search, email, display ads). Unlike traditional models, this method considers both the timing and type of interactions, making it uniquely suited to modern multi-touchpoint environments.
“Transformers give us the ability to see the journey as a whole, not just as a series of isolated interactions. That’s a major leap in marketing analytics.”
— PK Kannan
The core innovation lies in its use of multi-head self-attention mechanisms, which model each customer’s journey as a dynamic sequence of touchpoints. This allows marketers to not only predict the likelihood of purchase but also identify when and through which channels interventions are most effective. Furthermore, the model is extended to capture individual-level heterogeneity, enabling personalized insights into how different customers respond to marketing efforts.
“We designed the model to capture the complexity and individuality of digital customer journeys—something traditional models often overlook.”
— Zipei Lu
Using rich data from a major hospitality firm – including over 92,000 users and over half a million visits – the model demonstrates substantial improvements over traditional approaches (e.g., Hidden Markov Models, Poisson Point Process Models, and LSTMs). For example, the proposed model achieves an AUC of 0.92 for out-of-sample conversion predictions compared to 0.85 for LSTM and <0.70 for others. Moreover, it identifies high-potential customers with far greater precision – top-decile predictions yield an 88% true conversion rate versus 34% for LSTM.
Beyond prediction, the model offers descriptive marketing insights, such as how the effectiveness of email or display ads varies over time and across customers. For instance, the study finds that customer-initiated interactions (like direct visits) have stronger and longer-lasting effects than firm-initiated ones (like emails), and the optimal window for intervention is typically within 7 to 14 days before purchase.
The model’s structure also enables profiling and customer segmentation based on latent self-attention patterns, helping marketers understand nuanced motivations like last-minute business bookings versus long-term vacation planning. This insight can inform targeted messaging and A/B testing strategies.
Overall, this AI framework not only enhances predictive accuracy but also delivers actionable insights that can improve ROI, optimize channel mix, and enable real-time personalization in customer engagement.
Zipei Lu, Ph. D. Candidate in Marketing; P. K. Kannan, Dean's Chair in Marketing Science, both at the Smith School
From Man vs. Machine to Man + Machine: The Art and AI of Stock Analyses
Journal of Financial Economics, July 2024
An AI analyst trained to digest corporate disclosures, industry trends, and macroeconomic indicators surpasses most analysts in stock return predictions. Nevertheless, humans win “Man vs. Machine” when institutional knowledge is crucial, e.g., involving intangible assets and financial distress. AI wins when information is transparent but voluminous. Humans provide significant incremental value in “Man + Machine”, which also substantially reduces extreme errors. Analysts catch up with machines after “alternative data” become available if their employers build AI capabilities. Documented synergies between humans and machines inform how humans can leverage their advantage for better adaptation to the growing AI prowess.
Sean Cao, Robert H. Smith School of Business, University of Maryland
Logistics Service Provider Technology Report
Logistics Service Provider Technology Report
The Logistics Service Provider Technology Report (LSPTR) will be an annual report published by the University of Maryland’s Supply Chain Management Center that aims to provide technology spend visibility for logistics service providers (LSPs) in a variety of areas.
We find that LSPs do not know how much to invest in technology because public filings do not disclose specifics about IT spend, consulting firms have limited data to back their perspectives, and industry analysts are bias and do not collect hard data. Shippers also cannot compare providers' technology capabilities or investments due to LSPs alignment with strategy being unclear despite marketing various capabilities, and they cannot compare their partners’ technology investment within their segment or the broader market.
Publishing an annual technology report compiling technology spend data will provide a solution to the identified problems and create value for stakeholder groups including, but not limited to: LSPs, software vendors, hardware vendors, shippers, industry associations, trade groups, shareholders, and consulting firms.
The report will encompass all technology-related expenditures of the companies who opt in to provide a complete perspective of LSP interest, activity, and spend on technology, with an initial proof-of-concept/pilot addressing 2 key sub-sets of technology in 2025: AI and robotics.
Geoff Milsom - UMD Professor
Jaclyn Wilton - Advisor
Maggie McGuire - Fellow
Ryan Sachar - UMD Undergraduate Student
Ivy Zheng - UMD Undergraduate Student
AI-powered Analysts
We explore how brokerage firms’ investments in artificial intelligence (AI) affect their analysts’ information production. We find that analysts employed at brokerage firms with greater AI integration issue more accurate earnings forecasts. Cross-sectional analyses reveal that AI’s benefits are more pronounced for analysts with less firm-specific experience and when the firm’s disclosures are more readable. Further tests indicate that a key mechanism driving the improvement in forecast accuracy is that AI adoption helps mitigate the adverse effects of analyst decision fatigue and optimism bias. Finally, we find that forecast revisions made by AI-powered analysts are more informative to capital markets. Overall, our evidence points to the advantageous impact of AI on information production capabilities of financial analysts.
Michael Kimbrough, Musa Subasi, Liu Yang