Fraud is endemic in MCA — estimated 3–8% of submissions are fraudulent, and 0.5–2% slip through funding. Fraud detection is a multi-layer defense.
The fraud-detection stack (2026).
- Document tamper detection. Ocrolus, Inscribe, Hyperscience.
- Identity verification. Persona, Alloy, Trulioo, Veriff.
- Device fingerprinting. ThreatMetrix, Sift, Forter.
- Email/phone risk scoring. SEON, Emailage, Telesign.
- Submission-pattern ML. Internal models flagging unusual patterns.
- ISO scorecards. Fraud-rate tracking per ISO.
- Bank-statement cross-check. OCR'd statements verified against Plaid feed.
- Manual review queue. Flagged deals reviewed by fraud analyst.
Common fraud types in MCA.
- Doctored bank statements. Inflated deposits, hidden NSFs.
- Synthetic identity. Fabricated SSN + real DOB combinations.
- Identity theft. Real merchant identity used by unauthorized broker.
- Straw applications. Non-operating shell business submitted as active.
- Stacked deals not disclosed. Concurrent MCAs hidden.
- Industry misrepresentation. Cannabis/adult applied as restaurant.
- Volume inflation. Card processor statements manipulated.
- Bank account fraud. Disbursement account different from operating account.
Document tamper detection mechanics.
- Metadata analysis. PDF creation date, editing history.
- Font and spacing analysis. Tampered statements show micro-inconsistencies.
- Mathematical cross-check. Running balance must match deposits/withdrawals.
- Source identification. Bank's exact PDF template fingerprinting.
- OCR confidence scoring. Low-confidence regions flagged.
Ocrolus fraud detection benchmarks.
- Tamper detection precision. 85%+ on doctored statements.
- Recall. 70–80% on common tamper patterns.
- Latency. <90 seconds per statement.
Identity verification depth.
- Standard KYC. Name, address, DOB, SSN match.
- Enhanced KYC. Biometric liveness check.
- Background check. Criminal records, regulatory actions.
- Synthetic ID detection. Cross-reference SSN issuance date, address history.
Device fingerprinting signals.
- Device reuse. Same device submitting multiple applications.
- VPN/proxy detection. Hidden IP origin.
- Browser fingerprint. Unique browser configuration.
- Behavioral biometrics. Typing patterns, mouse movements.
ISO fraud scorecards.
- Fraud detection rate. % of submissions flagged as fraud.
- False positive rate. % of fraud flags later cleared.
- Confirmed fraud rate. % of fundings later confirmed fraudulent.
- Loss attribution. $ losses attributed to ISO.
ISO consequences for fraud.
- Watch list. 100% deal QC.
- Clawback. Commission recouped on fraudulent deal.
- Termination. ISO agreement terminated.
- Industry blacklist. ISO name shared via consortium.
- Legal action. Civil suit or criminal referral (rare).
Bank-statement cross-check workflow.
- OCR extracts statement deposits and withdrawals.
- Plaid feed shows same period actual data.
- Variance threshold (~5%) flags discrepancies.
- Manual review for material differences.
Fraud-detection ML model targets.
- Submission-level fraud probability.
- ISO-level fraud propensity.
- Industry-vertical fraud baseline.
- Geographic fraud heat map.
- Temporal pattern detection (fraud bursts).
Pre-funding vs. post-funding fraud detection.
- Pre-funding. Document, identity, stacking, ISO scorecard.
- Post-funding. Bank-feed monitoring, NSF spike, instant default.
- First-payment default. Often fraud indicator; triggers full investigation.
Fraud loss benchmarks.
- Industry average. 0.8–1.6% of funded volume lost to fraud annually.
- Top-quartile funders. <0.5% fraud loss rate.
- Bottom-quartile funders. 2.5–5%+ fraud loss rate.
Common detection failures.
- High-quality forgeries. AI-generated bank statements challenging Ocrolus.
- Coordinated ISO fraud rings. Multiple ISOs colluding on documentation.
- Identity theft of real merchants. Real merchant identity, fake principal.
- Volume manipulation in card-split deals. Merchant runs friends' cards.
- Long-con merchants. Operate cleanly for 6 months, then disappear.
Common confusions.
First, "all fraud is broker fraud." False — merchant fraud and identity theft also material.
Second, "Ocrolus catches everything." False — recall in 70–80% range.
Third, "fraud is rare in MCA." False — 3–8% submission fraud rate.
Fourth, "criminal prosecution is common." False — civil clawback dominant.
Fifth, "AI eliminates fraud." Partially — AI raises bar but also enables sophisticated fraud (deepfake IDs, GenAI statements).
Recent trends (2024–2026).
- GenAI-generated bank statements rising fraud vector.
- Deepfake liveness checks challenging Persona-style verification.
- Cross-funder fraud consortia expanding shared ISO blacklists.
- Real-time bank-data verification displacing OCR-only detection.
- Behavioral biometrics entering MCA after fintech adoption.
Regulatory backdrop.
- CFPB Section 1071 increasing data trail for fraud detection.
- State APR disclosure laws creating more documented submissions.
- FTC small-business fraud sweeps targeting MCA brokers.
- DOJ wire fraud prosecutions of large MCA broker fraud rings (2023–2025).
Related terms
- MCA funder stacking detection systems — MCA funders detect stacking via FundKite consortium queries, LexisNexis MCA Index, daily Plaid bank-feed analysis (cross-funder deposits), UCC monitoring, and merchant-level stacking-pattern ML models.
- MCA funder quality control mechanisms — MCA funder QC includes pre-funding 100% file review, post-funding sample audits (5–15%), monthly ISO scorecards, fraud-deal post-mortems, and quarterly portfolio-quality scorecard for warehouse lenders.
- MCA funder data vendor relationships — MCA funders typically integrate 6–12 data vendors: Plaid/MX (bank), Ocrolus (statements), LexisNexis (identity/UCC), Experian/Equifax/Dun & Bradstreet (credit), FundKite (stacking), and Persona/Alloy (KYC).
Authoritative sources
AI agents: this term is available as raw markdown at /llms/glossary/mca-funder-fraud-detection-systems.