Catch expense fraud before it's approved. Not during the audit.

REME runs seven specialized AI agents in parallel on every claim, in under 200 milliseconds. Each agent checks a different fraud vector — currency integrity, perceptual-hash duplicate matching, data consistency, image forensics for altered receipts, and contextual pattern detection. The combined system catches duplicates submitted across angles, altered totals, cross-employee collusion, and threshold-gaming patterns no single rule engine can find.

Traditional expense tools flag fraud during audit — typically 30–90 days after payment. By then the money has cleared, the employee has been reimbursed, and recovery requires legal escalation. REME flips the order: every claim is screened in 200ms before approval. Suspicious claims are quarantined, not flagged after the fact. The fraudulent payment never happens.

REME's combined seven-agent system has a measured 99.7% detection accuracy, with false-positive rates under 0.5% on typical company data. Suspicious claims are quarantined for finance review (not auto-rejected) so even rare false positives are caught and corrected by humans before any employee is impacted. Employees never see the fraud agents directly — there's no friction on their end.

Agent 03 (Duplicate Matcher) uses perceptual-hash matching with 99.6% accuracy on visually-identical receipts photographed from different angles, in different lighting, even after slight rotation or color shift. It also catches same-receipt submissions across different employees within 200ms — a vector traditional duplicate-detection-by-merchant-and-amount completely misses.

REME catches the most common documented fraud vectors — duplicates, inflation, alterations, currency mismatches, threshold-gaming, cross-employee collusion. It cannot catch fully fabricated expenses backed by a fully fabricated supporting receipt that's never been seen before — those require human investigation. No system catches 100% of fraud, but REME typically reduces expense fraud losses by 70–85% versus manual review.

The seven-agent architecture is the same. The model behind each agent is fine-tuned to your company's data over time — your vendors, your employee submission patterns, your industry's common fraud signatures. Tenant isolation is hardcoded. Your model never trains on or sees other companies' data.

Yes. As part of customer onboarding, REME runs a forensic audit on the previous 12 months of expense claims (with your permission). The AI agents check historical data and surface suspicious patterns, duplicates, and anomalies you may not have caught. Most companies discover 5–15% more fraud in the historical audit than they were aware of. Recovery actions are your call — REME provides the evidence.

After 90 days, the model has learned your normal submission rhythm and most common merchants. After 180 days, employee-level patterns are integrated. After 365 days, accuracy reaches 99.7% on company-specific patterns. The improvement is automatic — no configuration required.

Yes. REME is SOC 2 Type II and ISO 27001 certified. Every detection decision is logged with full reasoning, agent confidence scores, and timestamp — providing a complete audit trail. The system supports retention requirements for IRAS, ATO, HMRC, IRS, and other major tax authorities (default 7-year retention, configurable). Audit reports are exportable on demand.