Insights & Perspectives

Thoughts on AI implementation, chargeback management, fraud prevention, and building efficient operations.

📅 January 15, 2025 AI & Automation 5 min read

How AI Transformed Our Chargeback Win Rate (And Saved Us Millions)

When we integrated AI into our chargeback processes at Super.com, I didn't expect to see a 20% increase in win rates within the first quarter. Here's what worked, what didn't, and the lessons I learned about implementing AI in risk operations.

The Challenge

Processing thousands of chargebacks manually was unsustainable. Our team was spending 6+ hours per analyst per day just gathering evidence, crafting responses, and submitting disputes. The work was repetitive, error-prone, and frankly, burning people out.

The Solution

We implemented a two-phase AI approach:

  • Evidence Automation: AI pulled relevant transaction data, communication logs, and booking details automatically
  • Response Generation: Machine learning models analyzed winning disputes to craft optimized responses
  • Smart Routing: Cases were prioritized based on win probability and revenue impact

The Results

The impact was immediate and measurable. Within three months, we saw automated dispute win rates increase by 20%, and analyst workload decreased by 6 hours per day per person. More importantly, we recovered over $5M in revenue that year while scaling from 4,500 to 66,000 annual disputes.

Key Takeaways

Start with the repetitive tasks. Don't try to automate everything at once. Focus on the most time-consuming, rule-based processes first. For us, that was evidence gathering.

Human oversight is crucial. AI makes mistakes. We kept humans in the loop for high-value disputes and quality checks. This hybrid approach gave us the best of both worlds.

Measure everything. Track win rates, processing times, and revenue recovery before and after implementation. This data becomes your roadmap for continuous improvement.

📅 December 10, 2024 Fraud Prevention 4 min read

Pattern Recognition: How We Caught a $200K Fraud Ring

Sometimes the best fraud detection isn't about fancy algorithms—it's about knowing what normal looks like and spotting when something's off.

The Discovery

It started with a hunch. We noticed an unusual spike in chargebacks from Southeast Asia, all following a similar pattern: new accounts, high-value bookings, and disputes filed within 24 hours of travel.

The Investigation

Using BigQuery and SQL, I pulled transaction data across six months. The pattern became clear: coordinated accounts, similar booking behavior, and a network of shared payment methods and IP addresses.

The Impact

We built automated detection rules that flagged similar behavior in real-time, preventing over $200,000 in attempted fraud. More importantly, we shared these patterns with our fraud prevention team to strengthen platform-wide defenses.