Payments & Fraud

Chargeback Reduction at a Subscription Mobile App

13%→<1% Chargeback rate drop
ECM Exit Mastercard compliance achieved
Ethoca Dispute interception integrated

Problem

A privacy-focused mobile subscription app entered Mastercard's Excessive Chargeback Merchant (ECM) program after chargeback rates exceeded compliance thresholds. The processor was weeks from pulling the account.

Approach

Root cause analysis identified five systemic issues: Stripe Link bypassing CVC/AVS checks, repeat dispute cards not being blocked, default Stripe risk threshold too permissive, unprotected credit bundle purchases, and absence of chargeback alerts.

Outcome

Radar rule consolidation (35 rules reduced to 22 without reducing fraud coverage), 3DS gating on credit bundles, Ethoca integration for dispute interception, and formal compliance documentation accepted by the payment processor. ECM program exited successfully.

Related Service: Chargeback Rescue & ECM Exit →Read the Full ECM Exit Guide →Read the Visa VAMP Guide →
AI & Automation

AI Chat Widget Deployment

<2 Weeks Concept to production
1 Script Tag Webflow deployment
Vercel Edge API key kept secure

Problem

A consumer mobile app help center needed to evolve into a production AI experience without exposing API credentials and without rebuilding the existing Webflow site.

Approach

Support content was restructured into a semantically chunked JSON knowledge base with metadata tags for category, keywords, and source. A Claude-powered chat widget was built with Vercel Edge functions, a pattern-matching fallback layer for offline use, and embedded via a single script tag.

Outcome

Deployment from concept to production in under two weeks. The widget runs on the existing Webflow site without infrastructure changes and the Claude API key never touches the client's frontend.

Related Service: AI Automations →
Your Project

Ready to create the next case study?

Every engagement starts with a conversation. Let's discuss your fraud or AI challenge.

Book an Intro Call →