MarTech

Tracking System: 250M Monthly Requests, 263ms Latency

Challenge: ShareASale's tracking system was responsible for capturing every affiliate transaction—the source of the company's revenue. During platform consolidation following acquisition, the tracking system required modernization to maintain reliability at scale (250M monthly requests) while preserving accuracy and minimizing latency for real-time commission attribution.

← Back to Case Studies

Executive Summary

Designed multi-layer reverse proxy architecture enabling ShareASale’s revenue-critical tracking system to remain functional during platform consolidation while appearing unchanged to 250K+ publisher integrations. Built production system bridging two incompatible platforms using Cloudflare Workers + AWS Lambda, maintaining 263ms average latency at 250M monthly requests with 99.5% revenue protection and 75% error rate reduction.

Key Results:

  • 250M monthly requests handled seamlessly
  • 263ms average latency maintained
  • 99.5% revenue protection achieved
  • 75% error rate reduction through improved fault tolerance
  • Zero customer re-integration required

The Challenge

Mission-Critical System

Business Context: ShareASale’s tracking system was responsible for capturing every affiliate transaction—the source of the company’s revenue. During platform consolidation, this system needed to:

  • Continue operating through multi-year migration
  • Maintain 99.9%+ uptime (every dropped transaction = lost revenue)
  • Require zero customer re-integration (250K+ publisher sites)
  • Support 250M monthly tracking requests
  • Preserve backward compatibility with legacy parameters

Technical Reality:

  • ShareASale platform: Custom tracking URLs, parameters, cookie logic
  • Awin platform: Different URL structure, parameters, attribution model
  • Integration points: 250K+ publisher websites with hardcoded tracking pixels
  • Business constraint: Publishers cannot be asked to update their integrations

The Problem

Incompatible Platforms:

ShareASale tracking request:

https://shareasale.com/r.cfm?b=123&u=456&m=789&urllink=example.com

Awin platform expected completely different structure, parameters, and processing logic.

Cannot Change:

  • 250,000+ publisher websites have ShareASale tracking installed
  • JavaScript tracking pixels embedded across millions of pages
  • Deep links in emails, social media, and marketing materials
  • Asking publishers to re-integrate = massive customer disruption

Must Work:

  • Every tracking request must complete successfully
  • Attribution logic must remain consistent
  • Performance cannot degrade
  • Revenue cannot be impacted

The Solution: Multi-Layer Reverse Proxy

Architecture Overview

[Publisher Site] → [Cloudflare Edge]
                         ↓
                  [Routing Logic]
                         ↓
        ┌────────────────┴────────────────┐
        ↓                                 ↓
  [Awin Platform]                [ShareASale Legacy]
    (Primary)                         (Failover)

Architecture Layers:

Layer 1: Cloudflare (Edge)

  • WAF rules for traffic filtering
  • DDoS protection and rate limiting
  • Initial routing decisions based on rules
  • Custom Workers for intelligent request handling

Layer 2: Akamai (CDN)

  • Global traffic distribution
  • Caching for static content
  • Secondary failover routing
  • Geographic load balancing

Layer 3: AWS (Application)

  • Lambda functions for request transformation
  • Parameter translation ShareASale ↔ Awin
  • Retry and failover logic
  • Integration with Awin backend systems

Cloudflare Workers Implementation

Custom Routing Logic:

// Conceptual implementation
export default {
  async fetch(request) {
    const url = new URL(request.url);
    
    // Parse ShareASale tracking parameters
    const params = {
      merchant: url.searchParams.get('m'),
      affiliate: url.searchParams.get('u'),
      banner: url.searchParams.get('b'),
      destination: url.searchParams.get('urllink')
    };
    
    // Translate to Awin format
    const awinParams = translateParameters(params);
    
    // Route to Awin platform
    try {
      const response = await fetch(awinPlatformUrl, {
        method: 'GET',
        headers: buildHeaders(awinParams)
      });
      
      if (response.ok) {
        return response;
      } else {
        // Failover to ShareASale legacy
        return await failoverToLegacy(params);
      }
    } catch (error) {
      // Handle errors, retry logic
      return await handleError(error, params);
    }
  }
};

Key Features:

  • Parameter translation: ShareASale → Awin format
  • Intelligent routing: Primary (Awin) with failover (ShareASale)
  • Retry logic: Automatic retries on transient failures
  • Error handling: Graceful degradation
  • Logging: Request tracking for debugging

Cloudflare Services Used:

  • Workers: Edge compute for routing logic
  • KV: Configuration storage (routing rules, mappings)
  • D1: Request logging and analytics
  • Wrangler: Safe CI/CD deployment

AWS Lambda Integration

Awin-Side Processing:

On the Awin platform side, AWS Lambda functions handled:

  • Request validation: Ensure required parameters present
  • Data transformation: Convert ShareASale data structures to Awin
  • Attribution logic: Apply correct commission rules
  • Database writes: Store transaction data in Awin systems
  • Response generation: Return appropriate tracking pixel

Technology Stack:

  • Lambda: TypeScript functions
  • API Gateway: HTTP endpoint management
  • DynamoDB: Fast key-value lookups
  • RDS: Transaction storage
  • CloudWatch: Logging and monitoring

Integration Points:

  • Cloudflare Workers call AWS API Gateway
  • Lambda functions process and route internally
  • Awin backend systems receive standardized data
  • Response flows back through proxy layers

Observability & Monitoring

Comprehensive Monitoring:

DataDog:

  • Distributed tracing across all layers (Cloudflare → AWS → Awin)
  • Request latency percentiles (p50, p95, p99)
  • Error rate tracking by type
  • Custom dashboards for tracking health
  • Anomaly detection and alerting

CloudWatch:

  • Lambda execution metrics
  • API Gateway performance
  • RDS query performance
  • Cost tracking

OpsGenie:

  • Incident escalation and on-call rotation
  • Integration with DataDog alerts
  • Escalation policies by severity
  • Post-mortem workflow

Custom Dashboards:

  • Real-time request volume
  • Latency trends
  • Error rates by category
  • Failover frequency
  • Geographic distribution

Synthetic Monitoring:

  • Health checks every 5 minutes
  • End-to-end tracking request simulation
  • Alerting on failures
  • Geographic diversity (multiple regions tested)

Technical Challenges Solved

Parameter Translation

Challenge: ShareASale and Awin used completely different parameter structures and naming.

Solution: Built translation layer with comprehensive mapping:

const parameterMap = {
  // ShareASale → Awin
  'm': 'advertiser_id',
  'u': 'publisher_id', 
  'b': 'creative_id',
  'urllink': 'destination_url',
  'afftrack': 'tracking_reference'
  // ... dozens more parameters
};

Complexity:

  • 50+ parameters to map
  • Different data types (integers, strings, URLs)
  • Conditional logic (some parameters only in certain contexts)
  • Validation rules (ensure data integrity)

Challenge: ShareASale and Awin used different cookie structures for attribution.

Solution:

  • Read ShareASale cookies at edge
  • Translate to Awin cookie format
  • Set both cookie formats during transition
  • Gradually phase out ShareASale cookies

Considerations:

  • Cookie expiration handling
  • Domain scope differences
  • GDPR/privacy compliance
  • Cross-device attribution

High Availability

Challenge: 99.9%+ uptime requirement with zero tolerance for revenue loss.

Solution: Multi-Layer Failover

Primary Path: Publisher Site → Cloudflare → Awin Platform

Failover Path 1: If Awin fails → Route to ShareASale Legacy

Failover Path 2: If both fail → Return tracking pixel with retry queue

Retry Logic:

  • Exponential backoff for transient failures
  • Circuit breaker for persistent failures
  • Dead letter queue for manual recovery

Result:

  • 99.5% revenue protection (nearly perfect)
  • No single point of failure
  • Graceful degradation under stress

Results

Performance Metrics

Scale:

  • 250M monthly tracking requests
  • ~95,000 requests per hour average
  • ~1,600 requests per minute sustained
  • Peaks 3-5x average during holidays

Latency:

  • 263ms average response time
  • p50: 180ms
  • p95: 450ms
  • p99: 800ms

Availability:

  • 99.5% revenue protection (effectively 99.5% uptime)
  • Zero scheduled downtime
  • Unplanned outages < 4 hours annually

Reliability Improvements

Error Rate Reduction:

  • 75% error rate reduction vs. legacy system
  • Improved retry logic caught transient failures
  • Better error handling prevented cascading failures
  • Circuit breakers protected against overload

Fault Tolerance:

  • Multi-layer failover prevented revenue loss
  • Graceful degradation under stress
  • No single point of failure
  • Automatic recovery from transient issues

Business Impact

Revenue Protection:

  • 99.5% of transactions captured successfully
  • $100M+ annual revenue protected
  • Zero customer disruption required
  • Seamless migration from customer perspective

Customer Experience:

  • 250K+ publishers: Zero re-integration required
  • 10K+ merchants: Uninterrupted tracking
  • No change to user-facing experience
  • Maintained trust during complex migration

Migration Enablement:

  • Bought time for permanent Awin solution development
  • Allowed phased migration approach
  • Reduced migration risk significantly
  • Protected business continuity

Lessons Learned

What Worked Well

Multi-Layer Approach:

  • Defense in depth provided resilience
  • Multiple failover options prevented outages
  • Flexibility to iterate on routing logic

Edge Computing:

  • Cloudflare Workers = global low latency
  • Zero cold starts (vs. AWS Lambda)
  • Easy deployment and rollback
  • Cost-effective at scale

Comprehensive Monitoring:

  • Early detection of issues
  • Fast incident response
  • Data-driven optimization
  • Clear visibility for stakeholders

What I’d Do Differently

Earlier Load Testing:

  • Should have load tested at full scale earlier
  • Discovered some bottlenecks in production
  • More aggressive synthetic traffic testing

More Granular Alerts:

  • Initial alerts sometimes too noisy
  • Refined over time but could have been better day 1
  • More context in alert messages

Documentation:

  • Runbooks created reactively
  • Should have written comprehensive docs upfront
  • Knowledge transfer could have been smoother

What This Demonstrates

For Tracking & Attribution Roles:

  • Revenue-critical system design and implementation
  • High-availability architecture (99.5% revenue protection)
  • Cookie-based attribution logic
  • Cross-platform tracking integration

For Platform Engineering:

  • Distributed systems at scale (250M requests/month)
  • Multi-layer reverse proxy architecture
  • Edge computing with Cloudflare Workers
  • AWS Lambda serverless integration

For Site Reliability Engineering:

  • High-availability system design
  • Comprehensive observability (DataDog, CloudWatch)
  • Incident response and on-call processes
  • Performance optimization (263ms at scale)

For Technical Leadership:

  • Complex system integration across organizations
  • Risk management for revenue-critical systems
  • Stakeholder communication (business + technical)
  • Interim solution enabling long-term strategy

Technologies Used

Edge Layer:

  • Cloudflare Workers (JavaScript/Node.js)
  • Cloudflare KV (configuration storage)
  • Cloudflare D1 (request logging)
  • Cloudflare WAF (security)
  • Wrangler (deployment)

CDN Layer:

  • Akamai (global distribution)
  • Caching and routing rules

Application Layer:

  • AWS Lambda (TypeScript)
  • AWS API Gateway
  • AWS DynamoDB
  • AWS RDS (PostgreSQL)

Observability:

  • DataDog (tracing, metrics, logs)
  • CloudWatch (AWS monitoring)
  • OpsGenie (incident management)
  • Custom dashboards

Contact

Building revenue-critical tracking systems or complex platform integrations? Let’s discuss high-availability architecture, edge computing, and systems that protect business continuity.

Get in Touch: stevenleve.com/contact
LinkedIn: linkedin.com/in/steve-leve

← Back to Case Studies