Case Study: Global Ecommerce Brand

Fraud down, approvals up, trust calibrated

Even with a well-established fraud prevention tech stack, this global consumer brand was hitting a wall: review costs were rising, approvals were getting slower, and evolving threats kept slipping through. After implementing Elephant, they saw a 3BPS drop in chargebacks, 27% fewer manual reviews, and a 35% lift in model performance, in under 4 months.

The challenge

A global consumer brand already had a mature fraud operation. Known threats were flagged, and losses were controlled. But something wasn’t adding up.

Fraud was changing, and operational costs were climbing. Approval velocity had stalled altogether. Despite solid defenses, new threats were getting through, and trusted customers were still being slowed down. They didn’t need more rules. They needed a system that could recognize trust in real time.

Their goals were clear:

Reduce chargeback rates by 1–2 BPS

Cut manual reviews to reduce friction and cost

Expand approvals safely without opening the door to risk

Modernize decision systems to adapt faster to new fraud patterns

Where static systems broke down

Review volume kept climbing

Manual reviews ate up time and budget with too many borderline cases and not enough trust-based signals

Chargebacks weren't dropping

Even with aggressive rules, fraud still slipped through. Patterns were changing faster than logic could keep up

Trusted users were slowed down

High-trust signals like aged emails and low-risk IPs weren't being used, resulting in unnecessary friction

Rules only covered the known

The system caught what it was designed for, but new patterns still required too much manual work

How the system changed with Elephant

Elephant calibrated trust scores using over 430,000 historical transactions across a four month period, aligning signals with real fraud and approval outcomes. The model improved performance within weeks of deployment.

What we implemented:

A custom-calibrated trust model based on internal fraud, review, and approval history
Scoring rules built with predictive signals like phone line type, email/phone age, and IP risk
Recommendations for future lift, including integration of additional risk signals

elephant-case-studies-signal-intelligence

Proof of signal intelligence:

VOIP phone numbers: rejected in review 60.3% of the time
Aged emails and phones: 2x lower fraud risk than new ones
Trusted countries: up to 30x lower fraud rate than baseline regions

The real win? More trust and smarter decisions

This consumer brand didn't just close risk gaps, they built a more adaptive, resilient fraud program. By elevating trust as a signal and not just a side effect, they now make faster, smarter decisions grounded in real-time intelligence.

3BPS drop in chargebacks

Rules built on adaptive scoring and signal-level logic drove a 3 basis point reduction in chargebacks

27% fewer manual reviews

Trust scores enabled earlier risk separation, reducing the need for human triage and cutting review burden

35% uplift in scoring accuracy

Median scores for fraud vs. legitimate users separated cleanly, enabling smarter rule design