¿No posee una cuenta?
Expense Coding Syntax: Misclassification in AI-Powered Corporate ERPs
Agustin V. Startari.
AI Power and Discourse, vol. 1, núm. 1, 2025, pp. 1-10.

Resumen
This study examines how syntactic constructions in expense narratives affect misclassification rates in AI-powered corporate ERP systems. We trained transformer-based classifiers on labeled accounting data to predict expense categories and observed that these models frequently relied on grammatical form rather than financial semantics. We extracted syntactic features including nominalization frequency, defined as the ratio of deverbal nouns to verbs; coordination depth, measured by the maximum depth of coordinated clauses; and subordination complexity, expressed as the number of embedded subordinate clauses per sentence. Using SHAP (SHapley Additive exPlanations), we identified that these structural patterns significantly contribute to false allocations, thus increasing the likelihood of audit discrepancies. For interpretability, we applied the method introduced by Lundberg and Lee in their seminal work, “A Unified Approach to Interpreting Model Predictions,” published in Advances in Neural Information Processing Systems 30 (2017): 4765–4774.
To mitigate these syntactic biases, we implemented a rule-based debiasing module that re-parses each narrative into a standardized fair-syntax transformation, structured around a minimal Subject-Verb-Object sequence. Evaluation on a corpus of 18,240 expense records drawn from the U.S. Federal Travel Expenditure dataset (GSA SmartPay, 2018–2020, https://smartpay.gsa.gov) shows that the fair-syntax transformation reduced misclassification rates by 15 percent. It also improved key pre-audit compliance indicators, including GL code accuracy—defined as the percentage of model-assigned codes matching human-validated general ledger categories, with a target threshold of ≥ 95 percent—and reconciliation match rate, the proportion of expense records successfully aligned with authorized payment entries, aiming for ≥ 98 percent.
The findings reveal a direct operational link between linguistic form and algorithmic behavior in accounting automation, providing a replicable interpretability framework and a functional safeguard against structural bias in enterprise classification systems.
Texto completo
Dirección externa:

Esta obra está bajo una licencia de Creative Commons.
Para ver una copia de esta licencia, visite https://creativecommons.org/licenses/by-nc-nd/4.0/deed.es.
Para ver una copia de esta licencia, visite https://creativecommons.org/licenses/by-nc-nd/4.0/deed.es.