Real-time fraud detection. Explainable by design.
momentPay Shield scores every transaction in real time — combining a configurable rules engine, machine-learning models, sanctions and watchlist screening, and behavioural velocity analytics — and returns an approve, review or decline decision in under 10 milliseconds, without ever blocking the payment flow.
Shield
For teams who have outgrown static rule lists
Shield is built for risk and compliance teams who need a system analysts can tune themselves, that learns from confirmed outcomes, and that can prove — line by line — why it made every decision it made.
Four steps, one synchronous pass
Scoring runs in-line with the payment flow, so the decision is available before the transaction completes.
Transaction in
The transaction enters scoring synchronously, carrying the full context the decision will be built from.
Signals gathered
Rules, machine-learning models, sanctions screening and behavioural velocity features are evaluated in one chained pass.
Decision returned
Approve, review or decline comes back in under 10 milliseconds, with a numeric risk score behind it.
Case opened
If the transaction is declined or trips a configured rule, a case is created automatically, pre-populated with what fired.
Composable, ordered, and fail-safe
Each stage feeds the next, and the whole chain runs as a single pass. If any scoring component errors out, the transaction is never blocked — it passes through safely and the failure is logged.
- Sanctions screening runs first and can never be silently bypassed — even a fast-path allow rule is overridden the instant a sanctions match is detected.
- Gateway rules short-circuit early for instant allow-list bypasses or forced escalation, before full scoring cost is incurred.
- Activation rules evaluate in priority order, short-circuiting automatically on the first decline.
- Sanctions screeningOFAC, EU, UN and World-Check One
- Gateway fast-path rulesAllow-list bypass or forced escalation
- Velocity & behavioural featuresLive profile per card, account, merchant, device
- Derived risk indicatorsFailure rate, average ticket, custom formulas
- Rule evaluationPriority-ordered activation rules
- ML scoringAnomaly detection blended with fraud classifier
Six pillars behind every decision
No-code rules
A visual builder with AND/OR/NOT logic trees, rich comparators and custom formulas. Risk teams ship detection logic without an engineering ticket.
Machine learning
Unsupervised anomaly detection and a supervised fraud classifier, blended into one ensemble score and retrained automatically from confirmed outcomes.
Global watchlist screening
Fuzzy name matching against OFAC, EU and UN sources, plus integrated World-Check One for PEP and adverse-media checks.
Behavioural analytics
Live velocity profiles per card, account, merchant and device, with impossible-travel detection and IP risk classification.
Full explainability
A feature snapshot and fired-rule trail captured on every score, so any decision can be reconstructed line by line.
Case management
Flagged transactions become investigable cases with full rule and signal context attached, on workflows you configure.
Analysts change detection logic themselves
No engineering ticket, no deployment window. Every rule, function and script change propagates to the live scoring engine instantly, with zero downtime.
- Visual rule builder with AND/OR/NOT logic trees, not flat rule lists
- Comparators for equality, containment, numeric ranges, list membership, multiples and elapsed time
- Custom functions and derived fields defined in a guided, spreadsheet-like syntax
- Sandboxed inline scripts for advanced logic, with hard execution timeouts and no network, file or process access
- Reference dictionaries for BIN ranges, merchant categories and blocklists, maintained by analysts
Gateway rules
Fast-path conditions evaluated before full scoring, for instant allow-list bypasses or forced escalation.
Activation rules
The core decisioning layer, evaluated in priority order with automatic short-circuit on the first decline.
Live reload
Changes take effect on the running engine immediately — no restart, no redeploy.
Suppression windows
Temporarily mute a noisy rule or a known-good entity without turning off protection entirely.
Anomaly detection
An isolation forest flags statistically unusual transactions, even with no labelled fraud history to learn from.
Fraud classifier
A deep neural network learns directly from confirmed fraud and chargeback outcomes to catch known patterns.
Ensemble scoring
Both model outputs blend into a single, more reliable score rather than competing signals.
Feature importance
Every model reports which signals drove its score, in plain terms, for audit and analyst trust.
Learns from outcomes, not assumptions
Two complementary models run alongside the rules engine, and the system maintains them itself as new fraud labels arrive.
- Automatic retraining — the system watches for enough newly confirmed fraud labels and retrains without manual intervention
- Automated architecture search — training trials multiple network architectures and keeps the best performer
- Experiment tracking — data science teams compare model trials side by side with full accuracy, ROC and importance metrics before promoting a challenger
- Flexible label ingestion — confirmed outcomes feed back by file drop, SFTP or API, closing the loop
Screened on the transaction, and monitored after it
Transactions and onboarding entities are screened against sanctions and watchlist data both instantly and on an ongoing basis.
- Fuzzy name matching against OFAC, EU, UN and other configurable sources, resilient to misspellings, transliterations and partial matches
- Integrated World-Check One screening via LSEG for watchlist, politically exposed person and adverse-media checks
- Perpetual monitoring keeps a screened entity enrolled, surfacing new matches automatically
- On-demand due-diligence dossiers retrievable as full case reports for compliance records
- Automatic list refresh on a configurable schedule, with zero-downtime cache reloads
A sanctions hit is never bypassable
Fast-path allow rules exist so that known-good traffic clears without paying the cost of full scoring. That shortcut is overridden the instant a sanctions match is detected — no rule configuration can suppress it, and the match is recorded on the decision trail.
Behavioural & velocity analytics
Shield builds a live behavioural profile of every card, account, merchant and device rather than judging a transaction in isolation.
- Velocity counters over any window, from minutes to months, configured in the admin console
- Derived ratios such as failure rate and average ticket size, refreshed on every transaction
- Impossible travel detection across implausible location changes
- IP risk classification for datacentre and anonymisation exit-node traffic
Every decision is fully reconstructable
Critical for compliance reviews, cardholder disputes and model governance.
Feature snapshot
The complete set of signals considered at decision time is captured and stored, encrypted at rest.
Fired-rule trail
Every rule that contributed, and its exact configuration at that moment, recorded in human-readable form.
Dry-run verify mode
Run the exact production pipeline against a test payload on demand and get the full decision breakdown back.
Historical reprocessing
Replay archived transactions through updated rules to see how a change would have performed before it goes live.
Case management & investigation
- Automatic case creation on decline or on any custom rule trigger
- Cases pre-populated with transaction details and the exact rules and signals that fired
- Search and browse console with full context attached to every case
- Actions, statuses, macros and filters configurable per case type without code changes
Audit & governance
- Full audit logging of configuration changes, logins and administrative actions
- Decision snapshots encrypted at rest
- Model metrics retained per experiment for governance review
- Dry-run output usable as evidence for third-party integration certification
Enriches its own decisions
Shield does not operate in a silo. It reaches out mid-scoring for the data it needs, and plugs into the fraud-operations tooling you already run.
- External data enrichment — rules and models pull live data from third-party APIs such as device intelligence, geolocation and credit bureaus mid-scoring, with caching and latency guardrails
- Webhook notifications — configurable outbound alerts by email, SMS or HTTP fire automatically when conditions are met
- Pluggable into authorization — Shield serves the decision contract that muNSwitch's issuing side calls, so it drops into the authorization step without touching the network-facing layer
Enterprise-grade foundation
Multi-tenant by design
Each customer and business unit has fully isolated rules, models and data.
In-memory hot path
Rule and model lookups run off in-memory caches for sub-millisecond reads at high volume.
Zero-downtime changes
Rule, model and list updates propagate live across the running system instantly.
Encrypted at rest
Sensitive data, including full decision snapshots, is encrypted in storage.
See it on your own transaction data
We can run the dry-run verify tool against your payloads and walk through the full decision breakdown — the signals considered, the rules that fired, and the score behind the outcome.