Financial Services — Crypto
·Singapore·8 monthsFund Reconciliation and KYC Automation for a Crypto Exchange
Automated fund reconciliation and KYC verification processes for a cryptocurrency exchange, replacing manual teams with AI agents that process transactions continuously rather than in daily batches. The system enables near-instant fund availability while maintaining regulatory compliance with MAS guidelines for digital payment token services.
Near-instant
Fund availability (was next-day)
6 to 2
Reconciliation team size
Minutes
KYC processing (was hours-to-days)
Challenge
Cryptocurrency exchanges face a unique reconciliation challenge that traditional financial institutions don't: transactions occur 24/7 across multiple blockchains, each with different confirmation times, fee structures, and finality guarantees. This exchange processed deposits and withdrawals across more than a dozen blockchain networks, and each network's transactions needed to be reconciled against the exchange's internal ledger in near-real-time. A team of six analysts performed daily reconciliation — matching blockchain transactions to internal records, investigating discrepancies, and ensuring that customer balances were accurate. The daily batch reconciliation process created two problems. First, customers experienced delays in fund availability because deposits could only be credited after the day's reconciliation confirmed the blockchain transaction. In a market that moves 24/7 and where minutes can mean significant price differences, this delay was a competitive disadvantage. Second, the six-person reconciliation team was expensive and represented a scaling bottleneck — as transaction volume grew, the team would need to grow proportionally. KYC verification added another layer of manual work. The exchange was subject to MAS regulations for digital payment token services, requiring identity verification, sanctions screening, and ongoing monitoring for all customers. New account applications were processed manually, with analysts reviewing identity documents, running sanctions checks, and making approval decisions. Processing times for new accounts ranged from hours to days, causing customer drop-off during onboarding.
Approach
Reluvate deployed AI agents that perform continuous fund reconciliation across all supported blockchain networks. Rather than batch processing at end-of-day, the system matches blockchain transactions to internal ledger entries as they occur, using blockchain-specific confirmation thresholds to determine when a transaction is considered final. For each blockchain, the agent monitors pending transactions, tracks confirmations, and updates the internal ledger in near-real-time. Discrepancies are detected within minutes rather than the next day, and are automatically classified by type: timing differences (transaction confirmed but not yet processed), fee discrepancies (network fees differing from estimates), and genuine mismatches requiring investigation. The KYC automation system processes new account applications through an AI pipeline: document extraction (parsing identity documents from photos using computer vision), data validation (checking extracted data against submitted form data), sanctions screening (automated checks against OFAC, UN, EU, and MAS sanctions lists), and risk scoring (evaluating the applicant's risk profile based on jurisdiction, transaction patterns, and other factors). Low-risk applications with clean document verification and sanctions checks are approved automatically. Medium and high-risk applications are routed to compliance officers with the AI's analysis and a recommended decision. Ongoing monitoring was also automated. The system continuously screens the exchange's customer base against updated sanctions lists, monitors transaction patterns for suspicious activity indicators, and generates Suspicious Transaction Reports (STRs) for the compliance team's review when thresholds are triggered.
Design Notes
The reconciliation system was designed around the principle of blockchain-native confirmation logic. Each supported blockchain has a different confirmation model — Bitcoin uses probabilistic finality (6 confirmations is the convention), Ethereum has moved to proof-of-stake with different finality characteristics, and some chains offer instant finality. The system encodes each chain's finality model and uses it to determine when a transaction should be reconciled, rather than applying a one-size-fits-all approach. Change management for the reconciliation team was handled thoughtfully. Rather than eliminating the team, Reluvate repositioned them as the exception-handling and investigation layer. The AI handles the 95%+ of transactions that reconcile cleanly, while the analysts focus on the genuine discrepancies that require investigation and judgment. This reframing — from processing clerks to investigators — was well received and resulted in no involuntary role changes. Exception handling in crypto reconciliation must account for scenarios unique to blockchain: chain reorganisations (where confirmed transactions are reversed), smart contract interactions that produce unexpected state changes, and cross-chain bridge transactions that have complex reconciliation paths. For each exception type, the system has defined detection criteria, classification logic, and escalation paths. High-severity exceptions (potential loss of funds, potential compliance breach) generate immediate alerts to both the operations team and the compliance officer.
Result
Fund availability for customers improved from next-day to near-instant for deposits on supported blockchains. The six-person reconciliation team was restructured into a two-person investigation team handling only genuine exceptions and complex cases. KYC processing time for new accounts dropped from hours-to-days to minutes for standard applications, reducing customer drop-off during onboarding. The exchange maintained full MAS regulatory compliance throughout the transition to automated processing.
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.