Device intelligence
Device, session, IP and behavioural telemetry help assess the environment in which the request originates.
MINT / MINT SentinelDevice intelligence.
Device intelligence.
Transaction intelligence.
One view of risk.
MINT Sentinel evaluates device and transaction-level fraud risk before sensitive banking operations complete. It combines rules, graph intelligence and machine learning, with a continuing feedback loop to improve detection.
Discuss your use case →
Fraud can begin before a money transfer. Adding a beneficiary, changing credentials or performing a card operation can also require a closer look at the customer’s context.
Sentinel evaluates customer, device, session, IP, behaviour, transaction history and relationships in real time. It consolidates these signals into an overall risk score and recommends ALLOW, CHALLENGE or BLOCK with explainable reason codes.
Transaction-level checks
The requested action, history, beneficiaries and connected entities help reveal risk in the operation itself.
Tenant-specific rules, graph analysis and two complementary ML models contribute to the risk assessment.
Rules engine
Deterministic patterns, tenant-specific thresholds and risk-score adjustments.
Graph engine
Neo4j relationships help identify shared devices or IPs, suspicious clusters, known fraud links and potential mule activity.
Supervised learning
XGBoost learns known patterns from bank-confirmed fraud and legitimate outcomes.
Unsupervised learning
Isolation Forest identifies abnormal behaviour without requiring fraud labels.
Continual learning
Each confirmed outcome can improve the next assessment.
Sentinel’s learning loop uses feedback from confirmed outcomes to inform future detection. Supervised learning benefits from labelled fraud and legitimate cases; anomaly detection helps identify departures from normal behaviour.
The loop supports ongoing refinement as customer behaviour and fraud patterns change. Model and rule changes should follow the institution’s validation and release controls, rather than allowing unreviewed feedback to change live decisions.
The learning loop
Evaluate. Learn. Validate. Improve.
Evaluate
Assess the request and return risk evidence.
Confirm
Feed back the bank’s confirmed fraud or legitimate outcome.
Refine
Use feedback to inform model and rule improvements.
Validate
Assess changes before release, then monitor subsequent outcomes.
Sentinel is multi-tenant, with tenant-specific rules and thresholds. It provides intelligence for the decision; the bank chooses how to proceed with a sensitive operation.
The decisioning combines deterministic controls, graph intelligence and traditional supervised and unsupervised ML. It does not require a generative model to make the final banking decision.
Tell us about your customers, existing systems, operating markets and deployment needs. Our engineers work with your team to configure, customise and integrate MINT. Contact us →