Maher Kassar

Research

Publications

1. A Risk Assessment Framework for Cognitive Process Automation in Audit in Journal of Emerging Technologies in Accounting

   with Arion Cheong and Huaxia Li

   Abstract

   Cognitive process automation (CPA) is the process of automating knowledge-intensive tasks that require reasoning, interpretation, and decision-making using artificial intelligence agentic workflows. Although CPA offers significant potential in audit, auditors often struggle to determine what tasks are suitable for CPA and manage the risks associated with CPA implementation. This study proposes an artificial intelligence risk reporting tool that operationalizes an Artificial Intelligence Risk Assessment Framework for CPA deployment in auditing. Drawing on Task-Technology Fit Theory and Cognitive Load Theory, our framework includes three sequential stages: assessing audit task suitability for CPA, quantifying the cognitive load measure to proxy for artificial intelligence implementation risk, and translating the measured risk into prescriptive human oversight requirements. This study contributes to the literature by proposing a framework to bridge the current governance gap in CPA deployment and its associated risk.

2. Artificial Intelligence and Robotic Process Automation in Auditing and Accounting: A Systematic Literature Review in Journal of Applied Accounting Research

   with Mohammad Jizi

   Abstract

   This article provides a systematic review of the implementation of Artificial Intelligence and Robotic Process Automation (RPA) in the accounting and auditing professions. It uniquely and holistically identifies the benefits, challenges, drivers, and endorsement levels of implementing these emerging technologies and investigates gaps in the literature to prioritize future studies. The systematic review method was chosen to critically assess the body of literature following Tranfield et al. (2003) systematic review approach. The years 2016–2022 have been considered to gain insights into recent advancements in Artificial Intelligence and RPA deployments, and a rigorous selection process was designed to identify the relevant literature and ensure the quality and standard of the reviewed results. The study shows that the potential benefits of implementing Artificial Intelligence and RPA can be categorized into four categories—monetary, quality reporting, operational, and customer-related benefits—while the challenges and limitations can be categorized into four categories: ethical, regulatory, societal, and technical challenges. Initial studies generally show that accountants and auditors are not fully endorsing these implementations, mainly due to concerns about job loss and the trustworthiness of these technologies. The coding process of the examined articles was done manually, which could introduce subjectivity despite efforts to prevent this through multiple coders and rounds of review; the methodological approach used may also be criticized for eliminating articles with a journal ranking below 2. Moreover, the review only included articles written in English, which may exclude relevant studies, and it did not examine whether certain regions were over-represented in the sample—a limitation addressed by a recommendation in the future research directions section. The systematic review further shows that auditors, their clients, and regulatory bodies are working independently, which highlights the importance of a framework or mechanism to govern the relationship between them and build common expectations to scale up the utilization of RPA and Artificial Intelligence in a well-regulated profession; this review would help audit firms and organizations build strategies to better understand the vital factors and limitations towards scaling up the deployment of such technologies. Overall, this review can be seen as a guide for managers and policy makers and offers future research direction for scholars, as it facilitates (1) understanding which type of technology to utilize, given the different potential benefits of these emerging technologies in a well-regulated profession, and (2) realizing the factors holding back the potential of scaling up the adoption of Artificial Intelligence and RPA.

3. Supplier Concentration and the Speed of Capital Structure Adjustment in Pacific-Basin Finance Journal

   with Yi Liu, Kai Wu, and Sirui Ruan

   Abstract

   Using a large sample of A-share Chinese firms from 2012 to 2019, we show a positive relationship between supplier concentration and leverage adjustment speed. This positive relationship is unidirectional and primarily concentrated among over-leveraged firms rather than among under-leveraged ones. Moreover, the effect of supplier concentration is more pronounced in firms with greater bargaining power over suppliers. A plausible channel is the monitoring role imposed by suppliers, which mitigates the firm’s agency conflicts by reducing (1) information asymmetry between informed managers and uninformed market participants and (2) management opportunism and slacking. Further, we show that firms with higher supplier concentrations are more active in the security market due to having lower agency costs. Our findings demonstrate the crucial role of customer–supplier relationships in a firm’s capital structure dynamics.

Revise & Resubmit

1. Artificial Intelligence in Auditing: A Framework Linking Regulatory Requirements to Auditor Responsibilities in International Journal of Accounting Information Systems

   with Arion Cheong and James Biagi

   Abstract

   The rapid adoption of artificial intelligence (AI) by both auditors and their clients has introduced compliance obligations that existing frameworks were not designed to address. This paper examines how auditors ensure compliance with AI-related laws and regulations when AI is used within the audit engagement, whether deployed by the auditor or embedded in the client’s financial reporting processes. We conduct twelve semi-structured interviews with audit professionals, Chief Financial Officers, and AI providers, finding that auditors face a persistent diffusion of responsibility problem and lack unified guidance for navigating AI compliance obligations across both deployment contexts. Building on these findings, we develop a framework comprising four pillars—Accountability, Reporting, Risk Governance, and Security—each anchored to the COSO Enterprise Risk Management Framework and grounded in interview evidence, applicable AI regulations, and prior academic literature. Each pillar explicitly delineates auditor obligations depending on whether the auditor deploys AI or the client uses AI in financial reporting.

Working Papers

1. Decoding Voluntary Disclosure Decisions: An Explainable Machine Learning Approach

   with Joon Ho Kong and Mike Fu

   Abstract

   This paper proposes a machine learning (ML) approach to examine firms’ decisions regarding voluntary management earnings forecast disclosures. Utilizing advanced ML models and Explainable Artificial Intelligence (XAI), we find that firms’ voluntary disclosure decisions are largely influenced by external demand, proxied by sell-side equity analysts following, and litigation risk. In addition, our results show that firm performance and leverage also serve as strong predictors of forecast issuance. With respect to management forecast error, we find that firm performance variables and analyst following primarily influence management forecast error. Our results highlight the dual role equity analysts play in shaping firms’ information environment. Leveraging our best-performing predictive model, we construct our metric “Management Forecast Disclosure Expectation” (MFDE), which captures market expectations of firms’ future disclosure. We find that voluntary management forecasts that are unexpected relative to MFDE are associated with lower bid–ask spreads and bond spreads.

2. Does Socially Responsible Investor Attention Influence Firms’ ESG Commitment?

   with Arion Cheong, Joon Ho Kong, and Yu Gu

   Abstract

   This paper examines whether visible attention from institutional investors, affiliated with the Principles for Responsible Investment initiative, influences firms’ corporate social responsibility (CSR) performance. Drawing upon the catering theory and leveraging the 2015 public release of the SEC’s EDGAR log files, which made downloads of firm disclosures publicly traceable, we find that firms receiving attention from PRI signatories improve their CSR performance. Moreover, drawing upon instrumental stakeholder theory, we find that this relationship holds even when the attention originates from institutional investors with no equity holdings in the firm — a group we identify as foreseen stakeholders. These effects are observed among firms with fewer financial constraints that can promptly respond to the attention, and within the manufacturing industry with greater regulatory scrutiny. These findings hold in a battery of identification tests that address endogeneity.

3. Bug Bounty Program Adoption and Firms’ Cybersecurity Risk

   with Arion Cheong, Won No, and Maria Zhang

   Abstract

   Bug bounty programs (BBPs) engage external ethical hackers to identify and disclose software vulnerabilities. Using hand-collected BBP issuance data from HackerOne and Bugcrowd covering the 2014–2024 period, we investigate the determinants and implications of BBP adoption. We find that prior cybersecurity breaches, presence of a cybersecurity committee, negative cybersecurity-related news, and the SEC’s December 2023 Cybersecurity Rule are positively associated with BBP issuance. In addition, we find that BBP participation is negatively associated with subsequent cybersecurity breach likelihood, suggesting that BBPs are effective in strengthening firms’ cybersecurity risk management and vulnerability detection processes. Finally, we find that BBP participation is associated with less boilerplate language and greater numeric content in cybersecurity-related disclosures. These findings have implications for cybersecurity governance, disclosure regulation, and the operations management of vulnerability discovery.

4. Should Data Analytics and Artificial Intelligence Be an Integral Part of Cybersecurity Assurance?

   with Won No, Arion Cheong, and He Li

   Abstract

   The Securities and Exchange Commission (SEC) recently adopted a new rule on cybersecurity risk management and strategy, governance, and cyber incident disclosure by public firms. Specifically, public firms are now mandated to disclose the processes to assess, identify, and manage material cybersecurity risks while disclosing the cybersecurity risk assessment program in place. AICPA’s cybersecurity risk management reporting framework provides a robust framework for firms to communicate cybersecurity efforts and for practitioners to report on the management-prepared cybersecurity information. In this study, we introduce a seven-step process of using data analytics in testing cybersecurity controls while leveraging artificial intelligence to aid with executing and automating the proposed process. Using synthetic data, illustrative examples of the process are provided to demonstrate the role artificial intelligence in facilitating cognitive automation in cybersecurity risk assessment. This study demonstrates how foundation models can serve as flexible collaborators with practitioners to enhance the effectiveness of cybersecurity assurance. Finally, a set of current and critical challenges for applying artificial intelligence and data analytics in assurance engagement are also discussed.

5. From REA Diagrams to Synthetic Datasets: Leveraging Large Language Models to Enhance Accounting Education

   with Shanshan Yang, Arion Cheong, and Jiwon Nam

   Abstract

   This study examines how accounting instructors can leverage large language models (LLMs) to generate synthetic structured datasets using the Resources, Events, Agents (REA) ontology. Following a design science methodology, we develop a framework that leverages a custom GPT application and Chain-of-Thought prompting to generate synthetic datasets that mimic business processes and enable instructors to customize the datasets according to their learning objectives. To evaluate the application, we focus on the revenue cycle as an illustrative business process and assess 150 LLM-generated datasets. The results show that REA-guided prompting produces structurally compliant datasets that preserve economic duality and the cardinality constraints, while non-REA guided prompting does not reliably preserve these properties. In addition, datasets generated with instructor-specified learning objectives reliably contain predefined anomalies, while those generated without such objectives do not. Instructor and student surveys provide evidence that the framework is perceived as useful, easy to use, and capable of producing authentic datasets that support learning activities. The study contributes to the accounting education literature by providing a practical framework that enables instructors to generate realistic, customizable accounting datasets aligned with specific learning objectives.

Work in Progress

1. Decoding CECL Implementation by U.S. Banks: A Focus on Qualitative Adjustments and Artificial Intelligence Applications

   with Steve Yang, Arion Cheong, and Shanshan Yang

2. ESG Performance as a Mitigating Factor in the Relationship Between SEC Comment Letters and Audit Fees

   with Won No

Honors

Teaching Experience

Stevens Institute of Technology

References

Available upon request — will be sent soon.