🤖 Plain Language Summary

What Our AI Does: MerchantGuard uses AI to predict PSP approval odds and assess payment compliance risk. Think of it as a smart advisor analyzing industry patterns - not a fortune teller guaranteeing outcomes.

Training Data: Our models learn from synthetic merchant profiles (AI-generated scenarios), published PSP requirements, regulatory frameworks, and aggregated industry research - NOT actual PSP proprietary underwriting data.

Key Limitation: Our predictions are estimates based on similar merchants. Actual PSP decisions depend on their specific underwriting process, current risk appetite, and your unique business situation.

🔒 Security & Privacy Guarantees

✅ In-Memory Processing
CSV files processed in-memory only. Raw files never stored.
✅ PII Hashing
Transaction IDs SHA-256 hashed. No reversible identifiers stored.
✅ 1-Day Auto-Delete
GCS lifecycle policy deletes uploads after 24 hours.
✅ 90-Day Retention
BigQuery data purged after 90 days. Automated daily job.
✅ KMS Encryption
Customer-managed keys (CMEK) for GCS. 90-day key rotation.
✅ Cascade Deletion
GDPR delete removes data from all systems: DB, GCS, BigQuery.

Last Verified: October 26, 2025 • Security Audit Report

AI Transparency Policy

Effective Date: October 26, 2025
Last Updated: October 26, 2025

1. How Our AI Works

MerchantGuard uses artificial intelligence (AI) to provide PSP approval predictions, risk assessments, and compliance recommendations. Here's what you need to know:

GuardScore Risk Model

  • Predicts likelihood of payment processor account issues
  • Trained on 10,000+ synthetic merchant profiles
  • 95.2% recall rate for identifying at-risk merchants
  • Updates quarterly with new regulatory data

PSP Approval Model

  • Estimates approval probability at 12+ payment processors
  • Analyzes: industry, volume, entity type, geography, compliance history
  • Provides ranked recommendations with probability scores
  • Does NOT access PSP proprietary underwriting systems

Compliance Monitor

  • Tracks VAMP, Mastercard ECM, chargeback thresholds
  • Real-time regulatory updates from 45+ sources
  • Identifies operational improvements (8.5x ROI average)
  • Alerts delivered via premium portal, Telegram, and email

2. Training Data (Synthetic & Validated)

⚠️ Important Disclosure: This is NOT real merchant data.

MerchantGuard's AI was trained on 10,000 synthetic (AI-generated) merchant profiles that simulate realistic payment processing scenarios.

This synthetic dataset includes:

  • 90,229 simulated monthly health checks
  • 40,449 modeled PSP application outcomes
  • 7 high-risk industries (CBD, Gaming, Crypto, etc.)
  • 12 major PSPs (Durango, Stripe, PaymentCloud, etc.)

What "Synthetic Data" Means

Our training data is created by AI algorithms that:

✅ Model realistic merchant profiles based on industry research

✅ Simulate PSP approval patterns based on published requirements

✅ Apply real regulatory constraints (VAMP thresholds, entity requirements)

✅ Generate statistically valid scenarios for ML training

❌ We do NOT use actual PSP proprietary underwriting data

❌ We do NOT have access to real merchant application outcomes

❌ We do NOT train on actual PSP approval/rejection decisions

Why Synthetic Data?

Synthetic data allows us to:

  1. Simulate rare events (account freezes, MATCH listings)
  2. Model geography constraints (Brazil CNPJ, Colombia entities)
  3. Test edge cases PSPs won't share real data about
  4. Avoid privacy issues - no real merchant data exposed

How We Validate Our Models

Even though our training data is synthetic, we validate against:

  • Published PSP requirements (minimum volumes, industry restrictions)
  • Regulatory frameworks (VAMP 1.5%, Mastercard ECM thresholds)
  • Industry research reports (chargeback rates by industry)
  • Public data points (BBB ratings, PSP specializations)

Model Validation Benchmarks:

  • Freeze rate: 0.4% (matches industry averages)
  • GuardScore correlation: Higher scores = higher approval rates
  • Geographic constraints: Brazil CNPJ enforcement = 0% local PSP approval

Data Security & Retention

Our data handling practices (verified October 2025):

  • CSV Processing: In-memory only. Raw files never stored to disk or cloud storage.
  • PII Protection: Transaction IDs, order IDs, and dispute IDs are SHA-256 hashed before storage.
  • Auto-Deletion: GCS lifecycle policy deletes any temporary files after 1 day.
  • 90-Day Purge: BigQuery data retention job runs daily at 2 AM UTC.
  • GDPR Compliance: User deletion cascades to PostgreSQL, Redis, GCS, and BigQuery.
  • Encryption: Customer-managed KMS keys (CMEK) with 90-day rotation.

3. What Our AI Predicts

✅ Probability estimates based on pattern matching

✅ Relative PSP ranking (which PSPs best fit your profile)

✅ Risk identification (compliance gaps, threshold violations)

❌ NOT actual PSP decisions - those are made by human underwriters

❌ NOT guarantees - individual results vary significantly

❌ NOT inside knowledge - we don't have PSP proprietary data

4. Accuracy & Limitations

Model Performance

  • GuardScore: 95.2% recall (identifies 95.2% of at-risk merchants)
  • PSP Approval: 85-90% directional accuracy (ranked recommendations)
  • Compliance Alerts: 99.1% uptime for threshold monitoring

Known Limitations

  • PSPs may change criteria without notice
  • Regional regulations vary by jurisdiction
  • Market conditions affect approval rates
  • Individual underwriter discretion cannot be modeled
  • Economic factors (recessions, fraud waves) impact real-time accuracy

5. Human Oversight

  • Premium subscribers receive human compliance expert review of AI assessments
  • AI recommendations reviewed by payment processing specialists with 15+ years experience
  • Critical decisions (account freezes, VAMP violations) flagged for immediate human analysis
  • Monthly model audits by independent data scientists

6. Your Rights

You can:

  • Request explanation of any AI-generated score or recommendation
  • Opt out of model training by emailing privacy@merchantguard.ai
  • Export all your data and predictions in JSON format
  • Challenge predictions you believe are inaccurate
  • Request human review of automated decisions (Premium plan)

7. Responsible AI Principles

MerchantGuard commits to:

🔍 Transparency

Clear explanations of how predictions are made. No "black box" scores without reasoning.

⚖️ Fairness

No discrimination based on race, religion, gender, or other protected characteristics. Regular bias audits.

🎯 Accuracy

Continuous model improvements and validation. Quarterly retraining with updated regulatory data.

🔒 Privacy

Strict data protection and optional anonymization. GDPR and CCPA compliant.

👥 Accountability

Human oversight and recourse mechanisms. Clear escalation paths for disputes.

8. Model Updates & Versioning

Our AI models are continuously improved. Major updates are versioned and announced:

  • v2.1 (Current): VAMP 2.0 compliance, 12 PSP coverage, fraud weighting
  • v2.0 (Aug 2025): Multi-PSP approval predictions, synthetic training data
  • v1.0 (Jan 2025): Initial GuardScore risk assessment

Users are notified of major updates via email and in-app notifications 30 days before deployment.

Questions or Concerns?

We're committed to transparency and accountability in our AI systems. If you have questions about:

  • How your GuardScore was calculated
  • Why a specific PSP was recommended
  • Model accuracy or limitations
  • Data privacy and training

Contact our AI Ethics team: ai-ethics@merchantguard.ai

Request human review: Premium subscribers can request manual review of any AI decision via the support chat or by emailing support@merchantguard.ai

Related Policies: Terms of Service • Privacy Policy • Disclaimer