Payment Processing Architecture for High-Volume Merchants

The Payment Tax You Don't Know You're Paying

Most merchants treat payment processing as a commodity. Plug in Stripe, done. That works at $1M/year. At $10M+, the difference between a good and bad payment architecture is $200K-500K annually in declined transactions, unnecessary fees, and lost revenue.

The Three Problems at Scale

Problem 1: Decline Rates Are Higher Than You Think

Typical merchant decline breakdown:
  Successful:                     85%
  Soft declines (retriable):       8%
  Hard declines (card invalid):    4%
  Fraud blocks (false positives):  3%

That 15% failure rate on $10M = $1.5M in lost revenue.
Industry best practice: 92-95% success rate.
The gap is $200-500K in recoverable revenue.

Problem 2: You're Overpaying on Processing Fees

Standard Stripe pricing: 2.9% + $0.30 per transaction

On $10M annual volume:
  Total fees: $320,000

With optimized interchange routing:
  Qualifying for Level 2/3 data: 2.2% + $0.10 average
  Total fees: $230,000

  Savings: $90,000/year — just from sending better data

Problem 3: Single Point of Failure

When Stripe has an outage (and they do):
  - Every transaction fails
  - Cart abandonment spikes to 95%+
  - Revenue loss: $2K-50K per hour depending on volume

With multi-processor failover:
  - Primary processor down → automatic failover to secondary
  - Customer never notices
  - Revenue loss: near zero

The Architecture

Layer 1: Payment Orchestration

// Payment orchestration layer — route to optimal processor
interface PaymentRequest {
  amount: number;
  currency: string;
  cardBrand: string;        // visa, mastercard, amex
  cardCountry: string;      // issuing country
  isRecurring: boolean;
  merchantCategory: string;
  level2Data?: Level2Data;  // For B2B/corporate cards
}
 
function routePayment(request: PaymentRequest): ProcessorConfig {
  // Route based on optimization rules
  const rules: RoutingRule[] = [
    // Amex: Route to Amex direct (lower fees than via Stripe)
    { condition: (r) => r.cardBrand === "amex",
      processor: "amex_direct", reason: "Lower interchange" },
 
    // International cards: Route to processor with local acquiring
    { condition: (r) => r.cardCountry !== "US",
      processor: "adyen", reason: "Local acquiring reduces declines" },
 
    // Recurring/subscription: Route to processor with network tokens
    { condition: (r) => r.isRecurring,
      processor: "stripe", reason: "Network token support" },
 
    // High-value B2B: Route with Level 2/3 data for lower rates
    { condition: (r) => r.amount > 500 && r.level2Data,
      processor: "stripe", reason: "Level 2/3 data support" },
 
    // Default
    { condition: () => true,
      processor: "stripe", reason: "Default processor" },
  ];
 
  return rules.find(r => r.condition(request))!;
}

Layer 2: Smart Retry Logic

Not all declines are final. Smart retries recover 30-50%
of soft declines:

Soft decline (retriable):
  - Insufficient funds → Retry in 24h, then 72h
  - Processor timeout → Retry immediately on backup processor
  - Rate limit → Retry in 60 seconds
  - Generic decline → Retry once with different processor

Hard decline (do not retry):
  - Card stolen/lost → Do not retry, flag account
  - Invalid card number → Do not retry, request new card
  - Expired card → Send card update request, then retry

Network errors:
  - Timeout → Retry immediately on backup processor
  - Connection refused → Failover to secondary processor
  - 500 error → Retry with exponential backoff (1s, 2s, 4s)

Layer 3: Fraud Prevention That Doesn't Kill Conversion

The fraud paradox:
  Too strict: Block 5% of legitimate orders ($500K lost revenue)
  Too loose: Accept 2% fraud ($200K in chargebacks)
  Just right: Block fraud, approve good orders (requires nuance)

Layered approach:
  1. Device fingerprinting (passive, no customer friction)
  2. Velocity checks (3+ orders from same IP in 1 hour)
  3. Address verification (AVS match)
  4. 3D Secure (only for high-risk transactions)
  5. Manual review queue (for orders flagged but not blocked)

Key: 3D Secure should be ADAPTIVE, not universal.
  Low-risk orders: Skip 3D Secure (no friction)
  Medium-risk: Request 3D Secure (shifts liability)
  High-risk: Block and flag for review

Layer 4: Multi-Processor Failover

// Automatic failover when primary processor is down
async function processPayment(request: PaymentRequest): Promise<PaymentResult> {
  const processors = getProcessorPriority(request);
 
  for (const processor of processors) {
    try {
      const result = await processor.charge(request);
 
      if (result.success) return result;
      if (result.isHardDecline) return result; // Don't retry on hard decline
 
      // Soft decline — try next processor
      continue;
    } catch (error) {
      // Processor error — failover to next
      logFailover(processor.name, error);
      continue;
    }
  }
 
  return { success: false, reason: "All processors failed" };
}

Optimizing Interchange

Interchange fees vary by card type and data quality:

Standard consumer Visa:     1.65% + $0.10
Rewards consumer Visa:      1.95% + $0.10
Corporate Visa:             2.50% + $0.10
Corporate Visa (Level 2):   2.10% + $0.10  ← $0.40 savings per $100
Corporate Visa (Level 3):   1.90% + $0.10  ← $0.60 savings per $100

Level 2 data (send with every B2B transaction):
  - Tax amount
  - Customer reference number
  - Merchant postal code

Level 3 data (for large B2B transactions):
  - Line item details
  - Quantity and unit cost
  - Commodity codes
  - Ship-to postal code

Most merchants don't send Level 2/3 data.
Sending it costs nothing and saves 20-30% on B2B interchange.

The Dashboard

Daily Payment Health:

Authorization Rate: 93.2% (target: > 92%)
├── Approved: 93.2%
├── Soft declined (retried): 3.1% (recovered 1.8%)
├── Hard declined: 2.5%
└── Fraud blocked: 1.2%

Processing Costs:
├── Effective rate: 2.35% (down from 2.9%)
├── Savings from interchange optimization: $7,200 MTD
├── Savings from Amex direct routing: $2,100 MTD
└── Projected annual savings: $112,000

Failover Events:
├── This month: 2 events
├── Avg failover duration: 3.2 seconds
├── Revenue protected: $14,200
└── Customer impact: None detected ✅

Fraud:
├── Fraud rate: 0.08% (below 0.1% threshold)
├── Chargeback rate: 0.04% (well below 1% Visa threshold)
├── False positive rate: 0.6% (target: < 1%)
└── Manual review queue: 12 orders (avg resolution: 4 hours)

Implementation Priority

Month 1: Add smart retry logic
  → Recover 30-50% of soft declines
  → ROI: $40-80K/year on $10M volume

Month 2: Add Level 2/3 interchange data
  → Reduce processing fees on B2B transactions
  → ROI: $30-90K/year depending on B2B mix

Month 3: Multi-processor failover
  → Eliminate single point of failure
  → ROI: Prevents $2-50K per outage event

Month 4: Adaptive fraud rules
  → Reduce false positives while maintaining fraud prevention
  → ROI: $50-200K/year in recovered false declines

Payment infrastructure is the most under-optimized system in most e-commerce businesses. Every percentage point of authorization rate improvement at $10M volume is $100K in revenue. Every 0.1% reduction in processing fees is $10K/year. The math makes this the highest-ROI infrastructure investment you can make.