Selected work

Representative Papers

APEX-Accounting

Jul 29, 2026

This study evaluates the practical capabilities of state-of-the-art large language models on real-world accounting tasks, including reconciliation, accruals, journal entry preparation, and financial statement generation. To this end, we introduce the first high-quality, closed-domain benchmark tailored to accounting practice, comprising 10 synthetic enterprises and 160 expert-designed and scored tasks that support multimodal financial documents such as PDFs and spreadsheets, all evaluated under a fixed token budget. Our analysis uncovers a Simpson’s paradox between model performance and token consumption. While Claude-Fable-5 (Max) achieves the highest Mean Criteria@3 score at 56.4%, no model exceeds 21.5% on Pass@8, revealing substantial limitations in current models’ ability to perform complex accounting reasoning.

0 citationsRead paper

Payrolls to Prompts: Firm-Level Evidence on the Substitution of Labor for AI

Jan 28, 2026

This study addresses the lack of credible empirical evidence on the firm-level displacement of human labor by generative AI. Leveraging granular payment data from a major U.S. expense management platform and exploiting the release of ChatGPT as a quasi-natural experiment, the authors employ a difference-in-differences framework to examine shifts in firm expenditures between online labor and AI services. The paper provides the first direct causal evidence at the firm level that generative AI is partially substituting human labor. It introduces a novel exposure metric based on firms’ pre-existing share of online labor spending. Results indicate that highly exposed firms increased their AI expenditure share by 0.8 percentage points relative to less exposed firms by Q3 2025, with every $1 reduction in online labor spending associated with an average $0.03 increase in AI spending, revealing significant cost-saving potential.

0 citationsRead paper

Recent publications

Latest Papers

APEX-Accounting

Jul 29, 2026

This study evaluates the practical capabilities of state-of-the-art large language models on real-world accounting tasks, including reconciliation, accruals, journal entry preparation, and financial statement generation. To this end, we introduce the first high-quality, closed-domain benchmark tailored to accounting practice, comprising 10 synthetic enterprises and 160 expert-designed and scored tasks that support multimodal financial documents such as PDFs and spreadsheets, all evaluated under a fixed token budget. Our analysis uncovers a Simpson’s paradox between model performance and token consumption. While Claude-Fable-5 (Max) achieves the highest Mean Criteria@3 score at 56.4%, no model exceeds 21.5% on Pass@8, revealing substantial limitations in current models’ ability to perform complex accounting reasoning.

0 citationsRead paper

Payrolls to Prompts: Firm-Level Evidence on the Substitution of Labor for AI

Jan 28, 2026

This study addresses the lack of credible empirical evidence on the firm-level displacement of human labor by generative AI. Leveraging granular payment data from a major U.S. expense management platform and exploiting the release of ChatGPT as a quasi-natural experiment, the authors employ a difference-in-differences framework to examine shifts in firm expenditures between online labor and AI services. The paper provides the first direct causal evidence at the firm level that generative AI is partially substituting human labor. It introduces a novel exposure metric based on firms’ pre-existing share of online labor spending. Results indicate that highly exposed firms increased their AI expenditure share by 0.8 percentage points relative to less exposed firms by Q3 2025, with every $1 reduction in online labor spending associated with an average $0.03 increase in AI spending, revealing significant cost-saving potential.

0 citationsRead paper