Compensatory project-score changes in an AI-supported financial management curriculum: course-embedded evidence and self-reported transfer
Year: 2026
Type: article
Abstract: Amid the current wave of artificial intelligence, education is entering a period of experiment in which teaching and generative AI are being combined, and some business schools are now piloting ways of bringing generative AI into their curricula. The course-level evidence, however, remains thin on whether AI-assisted learning builds real analytical capability. Teachers worry in two directions at once: that AI will widen existing learning gaps and deepen inequality among students, and that students may not gain any real skill from AI-assisted learning at all. This study contributes the answerable, distributional piece of both questions: when AI support is embedded in course design, can students who start with weaker quantitative and computational project performance enter analytics workflows more effectively? We examine a two-semester Financial Management curriculum reform that bundled Python, Jupyter Notebook, authentic business projects, staged assessment, instructor guidance, and guided generative-AI use with explicit student responsibility; the bundle cannot isolate the effect of AI alone. Evidence comes from two survey waves (n = 81, n = 79) and official course records: grade records for 90 students and a matched project-score sample of 73. The central theoretical contribution is compensatory AI scaffolding: AI support may be educationally useful when authentic tasks, reproducible notebooks, and assessment requirements return responsibility for checking and interpretation to students. In course-embedded project scores the Low–High baseline gap narrowed from 15.3 to 3.1 points; the low, middle, and high terciles changed by 13.1, 7.1, and 0.9 points, and a student fixed-effects model estimated 12.18 additional points of change for lower-baseline students (8.09 points under the more conservative final-exam stratification). With near-universal AI adoption, score changes were unrelated to individual AI-use intensity, so the pattern is not a product of heavier AI reliance among lower-baseline students. The study remains a course-based before-after design without random assignment, a concurrent control group, or externally standardized assessment, and because both semesters were graded by the same instructor, convergence in applied grading standards cannot be excluded. Matched-panel perception changes were directionally positive but not significant after false-discovery-rate correction and are reported descriptively.
Source: Management & Marketing
Authors: Yijiao Liu, Wenhao Chang
Institution: Shenzhen Technology University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $323
Funder: Shenzhen Technology University
OpenAlex ID: W7203832196