Alert fatigue
Static rules generated huge false-positive volume, so genuine fraud signals were lost in the noise.
Case StudyFintechRisk & Compliance
Fraud Detection & eKYC
Real-time fraud detection, AML monitoring and eKYC for a digital bank.
ProjectOverview
As account growth accelerated, so did fraud attempts and compliance load. Zimozi built a real-time risk engine and a streamlined eKYC onboarding flow that cut fraud losses and false positives at once, while keeping the bank aligned with MAS AML/CFT obligations.
Industry: Digital Bank · Region: Singapore (MAS-regulated) · Services: Fraud/AML Engineering, eKYC, Data & ML · Engagement: 9 months. Client identity withheld under NDA; figures representative of the engagement outcome.
TheChallenge
The bank's onboarding was slow enough to lose applicants, while its fraud controls were rule-based and noisy: thousands of low-value alerts drowned the real threats. With transaction volume climbing, the bank needed to demonstrably meet MAS AML/CFT and customer due-diligence expectations.
Onboarding drop-off
Manual, multi-step KYC caused applicant abandonment and slow time-to-account.
Slow to adapt
New fraud patterns took weeks to encode as rules, always a step behind the fraudsters.
Compliance evidence
Hard to produce the audit trail, screening and reporting MAS AML/CFT obligations require.
The Solution
We score every transaction in real time, and explain why.
A real-time scoring service evaluates each transaction in under 200ms using behavioural features, device signals and velocity checks, blending ML models with a transparent rules layer.
Risk-tiered alerts, entity/behaviour profiling and sanctions/PEP screening mean analysts see prioritised, explainable cases instead of undifferentiated noise, alongside a frictionless eKYC flow with digital identity, document and liveness verification.
Every decision carries its reason codes and a full audit trail, and models are monitored for drift, built to evidence MAS AML/CFT compliance with manual review reserved for genuine edge cases.
KeyFeatures
A single, prioritised queue where the few cases that matter surface first.
01
Streaming
Streaming
risk engine
Real-time scoring in under 200ms, blending ML models with a transparent rules layer.
02
AML transaction
AML transaction
monitoring
Risk-tiered alerts, entity profiling and sanctions/PEP screening.
03
Frictionless
Frictionless
eKYC onboarding
Digital identity, document and liveness verification with automated CDD checks.
04
Explainability
Explainability
& audit trail
Every decision carries reason codes and a full, MAS-aligned audit trail.
05
Model drift
Model drift
monitoring
Continuous monitoring keeps risk models accurate as fraud patterns evolve.
TechStack
Real-time stream processing, ML risk scoring, a feature store, sanctions/PEP screening and Singpass-style digital identity, aligned to MAS AML/CFT and PCI-DSS.
Python
AWS
MongoDB
Business Impact
Less fraud, less friction, less noise.
Fraud loss dropped 40%+ within two quarters, false-positive alerts fell 60% versus the rules-only baseline, and median eKYC onboarding is now ~3 minutes, fully digital, with an audit-ready trail for every decision.
Get started