Financial Services — Payments
·Singapore·9 monthsAI Fraud Detection and Reconciliation for a Regional Payments Company
Built an AI-powered fraud detection and merchant reconciliation system for a regional payments company processing transactions across multiple payment rails. The system analyses transaction patterns in real-time to flag fraudulent activity, automates merchant settlement reconciliation, and generates regulatory compliance reports for MAS reporting requirements.
62%
Reduction in fraud losses
8 to 3
Reconciliation team restructured
Automated
MAS regulatory reporting
Challenge
Payment processors operate on razor-thin margins where fraud losses and operational inefficiency have an outsized impact on profitability. This regional payments company processed transactions across credit card rails, e-wallet integrations, QR code payments, and bank transfer APIs — each with different settlement timescales, fee structures, and reconciliation requirements. The operations team manually reconciled merchant settlements daily, matching transaction records from payment gateway logs against bank settlement files and merchant statements. With thousands of merchants and tens of thousands of daily transactions, the reconciliation process consumed a team of eight and was still prone to errors. Fraud detection was reactive rather than predictive. The company relied on rule-based filters — transaction amount thresholds, velocity checks, geographic restrictions — that caught obvious fraud but missed sophisticated patterns. Card testing attacks, merchant collusion schemes, and account takeover fraud slipped through because the static rules couldn't adapt to evolving fraud tactics. Each fraud incident not only caused direct financial loss but also triggered chargeback fees, scheme fines, and reputational damage with acquiring banks. Regulatory reporting added another layer of manual work. As a MAS-licensed major payment institution, the company was required to submit periodic reports on transaction volumes, suspicious activity, and compliance metrics. These reports were assembled manually from multiple data sources, a process that consumed compliance team time and carried the risk of reporting errors.
Approach
Reluvate deployed a three-layer system covering fraud detection, reconciliation automation, and compliance reporting. The fraud detection layer analyses transactions in real-time using ML models trained on the company's historical transaction data, including both confirmed fraud cases and false positive patterns. The models evaluate each transaction against dozens of features — amount patterns, merchant category, time-of-day, device fingerprint, behavioural biometrics, and cross-merchant velocity — producing a risk score within milliseconds. High-risk transactions are blocked or held for review; medium-risk transactions are flagged for post-transaction monitoring. The reconciliation engine automates daily merchant settlement matching across all payment rails. The system ingests transaction logs from the payment gateway, settlement files from acquiring banks, and fee schedules per merchant agreement. It performs three-way matching — gateway record to bank settlement to merchant statement — identifying discrepancies categorised by type: timing differences (settled but not yet reported), fee calculation variances, duplicate transactions, and genuine mismatches requiring investigation. The operations team now reviews only the exceptions rather than processing every transaction. Compliance reporting is automated through scheduled data aggregation and report generation aligned with MAS reporting templates. The system produces transaction volume summaries, suspicious transaction logs, and compliance metrics in the formats required by MAS, ready for the compliance officer's review and submission. Ad hoc regulatory inquiries can be answered through a query interface that searches across the full transaction history.
Design Notes
The fraud detection models were designed to minimise false positives, which are operationally more costly than they appear. Every false positive — a legitimate transaction flagged as fraudulent — creates a negative customer experience, generates merchant complaints, and consumes analyst time for investigation. Reluvate tuned the models to optimise for the business's specific cost function: the ratio of fraud loss prevented to legitimate revenue disrupted. This required close collaboration with the company's risk team to quantify the actual cost of each outcome. Change management for the operations team focused on the reconciliation workflow. The team of eight had developed highly specific manual processes and institutional knowledge about common exception patterns for different merchant categories and payment rails. Reluvate captured this knowledge during a structured knowledge transfer phase, encoding it into the reconciliation engine's exception classification rules. The team transitioned from end-to-end reconciliation to exception investigation, which required different skills and was positioned as a more analytical, higher-value role. Exception handling in payment reconciliation must account for the asynchronous nature of multi-rail settlement. Different payment methods settle on different timescales — card transactions may take T+1 or T+2, bank transfers may be same-day, and e-wallet settlements follow platform-specific schedules. The system maintains a temporal matching window per payment rail and only escalates discrepancies that persist beyond the expected settlement window, avoiding false alarms from timing differences.
Result
Fraud detection rates improved significantly while false positive rates decreased, saving analyst time and reducing legitimate transaction disruptions. The reconciliation team was restructured from eight to three analysts focusing on complex exceptions and merchant relationship management. Regulatory report generation shifted from a multi-day manual process to an automated output requiring only compliance officer review and approval. The company's payment processing margins improved measurably as fraud losses declined and operational costs decreased.
Related Case Studies
Financial Services & Regulation
AI Sustainability Reporting for a National Financial Regulator
Built an AI-driven sustainability reporting framework for a multi-stakeholder consortium led by the Monetary Authority of Singapore (MAS), involving a sovereign wealth fund and Tier 1 international banks. The system automates the aggregation, analysis, and generation of regulatory sustainability reports that previously required months of manual effort across participating institutions.
Professional Services
Multi-Jurisdiction AI Accounting for a Corporate Services Firm
Deployed LynxLedger — an AI accounting system — for a corporate services firm managing hundreds of client entities across five countries. The system executes bookkeeping, reconciliation, and financial reporting autonomously while respecting jurisdiction-specific accounting standards including SFRS, HKFRS, MFRS, AASB, and NZ IFRS.
Healthcare
Predictive Financial Intelligence for a Multi-Location Healthcare Group
Built a real-time financial intelligence platform for a medical group operating 40+ clinic locations across three clinical segments. The system pools financial data from all locations, generates AI-authored commentary on performance and variance, and uses SARIMA forecasting models to predict revenue by segment.