Zarelva · Fraud Intelligence Framework

Decision-Layer Risk Engine

How raw platform events become fraud decisions — and how they continuously improve over time.

Read left to right, top to bottom·Data flows in·Signals are built·Engine decides·Humans tune

Identity

Device

Behaviour

Transaction

Network

Decision Engine

Live signal flow

01 · Platform & Data Inputs

🪪 Onboarding & KYC Identity, docs, bureau

📱 Logins & Devices Sessions, fingerprints

💸 Transactions & Payments UPI, BNPL, disbursements

⚠️ Disputes & Collections Chargebacks, NPA signals

🤖 AI & Agent Actions Autonomous flows, LLM calls

Raw events from your app — every touchpoint that carries a fraud signal

transforms into signals

02 · Layered Fraud Signal Library

01Identity & KYC
              document mismatch
              synthetic ID patterns
              KYC vendor risk codes
              bureau thin file
            
02Access & Device
              device fingerprint
              IP / geo mismatch
              emulator / root signals
              session anomalies
            
03Behaviour & Velocity
              rapid-fire applications
              typing / mouse patterns
              time-on-form outliers
              repeat failure loops
            
04Transaction & Monetary
              amount outliers
              merchant / category risk
              repayment history gaps
              disbursement timing
            
05Network & Ecosystem
              shared devices / IPs
              shared bank accounts
              link analysis rings
              mule cluster signals
            

Signals that actually describe fraud risk — not just raw data

feeds into engine

03 · Decision-Layer Risk Engine

Core Decision Engine

Decision-Layer Risk Engine

📏

Rules EngineHard cutoffs, threshold logic, conditional blocks

📊

ScorecardsWeighted signal combination, composite risk scores

🧠

ML Models (optional)Pattern detection, anomaly scoring at scale

📋

Override Logic & Audit TrailEvery decision documented and defensible

Agent Module

Agent Risk Engine

Prompt injection detection

Delegation chain abuse

Non-human velocity signals

Autonomous flow overrides

Attaches to Behaviour & Network layers. For AI-agent and autonomous decision flows.

Where decisions really happen — approve, review or decline logic lives here

produces outcomes

04 · Decision Outcomes

✅ Low Risk Approve

Signals below threshold. Transaction clears automatically. No friction added.

🔍 Needs Review Review

Signals elevated. Routed to fraud team or step-up verification before proceeding.

🚫 High Risk / Confirmed Fraud Decline · Block · Escalate

Signals confirm fraud. Transaction blocked, account flagged, case escalated for investigation.

Human Feedback Loop

Fraud & Risk Team
Zarelva FPaaS

Reviews outputs, tunes rules, and feeds new patterns back into the engine.

📄 Weekly Fraud Summary Every Monday · 12:00 IST

Incident count, pattern shifts, watchlist updates, top attack vectors.

📊 Monthly Insight Memo Monthly · PDF + call

Loss trends, rule decisions, top patterns. Board-ready language.

🔧 Rule & Scorecard Tuning Ongoing · version-controlled

Recommendations on what to add, retire, tighten. Living backlog.

🏗️ Architecture Review Quarterly · board report

Controls, data flows, gap analysis. Suitable for investors & auditors.

Feeds back into engine

New fraud patterns

Rule changes & additions

Threshold adjustments

Architecture updates

↑ continuous feedback ↑

Apply This Framework

Is your decision layer built like this?

Most digital lenders have signals but no architecture connecting them. Zarelva reviews your current setup and shows you the exact gaps — in 48 hours.

Run a Fraud Risk Assessment → Book a 30-Min Call

$599 · Fixed fee · NDA before any data is shared · Delivered in 48 hours