Banking fraud platform pilot

Context

Retail bank UZ. Card fraud rising 25% YoY. Social engineering cases growing — customers transferring money to fraudsters, public scandals. Existing fraud detection — fragmented per category (card processor, online banking, internal).

CRO task — consolidated fraud loop.

Diagnostic (4 weeks)

Fraud loss assessment:

• Card fraud: 0.04% transaction volume (industry average 0.05).
• Social engineering: estimated $1.2M annualised losses, growing.
• Account takeover: rarely detected, likely under-detected.
• Cross-category cases: invisible.

Approach (8 months)

Months 1-2. Architecture design. Customer-centric fraud platform consolidating signals across channels.

Months 2-4. Phase 1 — card fraud upgrade. Real-time scoring engine, ML model retraining, false positive tuning.

Months 4-6. Phase 2 — social engineering detection. Behavioural patterns + step-up authentication for suspicious transfers.

Months 6-8. Phase 3 — cross-category investigation workspace. Fraud analysts work with unified case context.

Results

After 8 months:

• Card fraud loss reduction: 42%.
• Social engineering cases prevented: 89 cases, ~$680k saved.
• False positive rate −33% — better customer experience.
• Investigation throughput +60% per analyst.
• Customer NPS post-fraud-block (false positives): +14 points (faster resolution).

What is critical

Customer-centric approach — not “card fraud” vs “online fraud”.

Step-up authentication critical for social engineering.

Customer feedback loop in real time — confirms / denies suspicion for model training.

Related

• /en/solutions/banking-fraud-platform/ — fraud platform
• /en/architecture/banking-realtime-decisioning/ — decisioning
• /en/use-cases/banking-suspicious-transaction-stepup/ — step-up
• /en/expertise/banking-fraud-discipline/ — fraud expertise