Natural Language Processing-Based Layered Reconciliation System for Financial Transaction Analysis

With the widespread adoption of digital payment systems, the volume and diversity of financial transaction data have increased significantly. For payment institutions and electronic money companies in particular, the cross-verification of internal transaction logs with bank statements has become a critical requirement for ensuring financial security, accounting accuracy, and auditability. However, in practice, inconsistencies often occur between bank-side and firm-side records due to system interruptions, service errors, or manually entered transactions. This study presents a financial data reconciliation system based on Natural Language Processing (NLP) and rule-based analytical techniques, designed to detect inconsistencies by comparing bank transaction records with internal operational logs. The system, developed by Elekse, automatically retrieves millions of transaction records from multiple banks via the Finekra platform and classifies them by transaction type using key attributes such as description, date, amount, and IBAN. Throughout this process, NLP techniques are used to identify linguistic patterns, extract meaningful expressions, and assign the appropriate accounting codes through predefined rules, enabling the automatic reconciliation of records. As a result, the need for manual inspection is reduced, error detection is accelerated, and overall data accuracy is improved.