Smart Remittance Router — AI-Powered Rail Selection
Intelligent routing engine: partner scoring → real-time health → cost optimization → success prediction → dynamic routing · maximizes delivery speed & minimizes cost
1 Partner & rail data collection
Partner registry
Thunes
Wise
Flutterwave
Cellulant
Circle (USDC)
Real-time metrics
Success rate (24h)
Avg delivery time
Current latency
Queue depth
Error rate
Updated every 30s
Historical performance
7-day success trend
30-day avg speed
Failure patterns
Peak hour behavior
Seasonal trends
Cost data
Base fee per partner
FX markup (partner)
Volume tier discounts
Minimum charges
Hidden fees (if any)
Corridor mapping
US → Nigeria (5 rails)
US → UK (4 rails)
US → Ghana (3 rails)
US → Kenya (4 rails)
50+ corridors total
2 AI scoring model
Feature extraction
Transaction amount
Send/receive currency
Recipient type (bank/wallet)
Time of day (local)
Day of week
Success prediction
XGBoost classifier
Per-partner probability
Confidence interval
Failure reason pred
Retry likelihood
Model accuracy: 94.2%
Speed prediction
Estimated delivery (mins)
P50 / P90 / P99
Queue wait time
Processing time
Settlement lag
Cost calculation
Total fee estimate
FX cost (spread)
Partner markup
Kredete margin
Net cost to user
Composite score
Speed weight: 30%
Cost weight: 25%
Success weight: 35%
Reliability weight: 10%
Final rank per rail
3 Dynamic route selection
Constraint filtering
Amount within rail limits
Recipient type supported
Partner currently healthy
Compliance rules pass
Liquidity available
User preference override
Fastest (pay premium)
Cheapest (may be slower)
Balanced (default)
Specific rail (power user)
Avoid partner (if prev fail)
Multi-path evaluation
Direct path (A→B)
Bridge path (A→USDC→B)
Multi-hop (A→C→B)
Split send (2 partners)
Fallback chain order
Final selection
Primary rail selected
Backup rail #1
Backup rail #2
Auto-failover enabled
Route ID logged
Optimal route rate: 98.1%
4 Execution & real-time monitoring
Transaction dispatch
Partner API call
Request signing
Idempotency key
Timeout config
Retry policy
Status tracking
Submitted
Processing
In transit
Delivered
Failed
Failure handling
Error classification
Auto-retry (transient)
Failover to backup rail
Manual review queue
Refund trigger
Webhook handling
Partner status callback
Delivery confirmation
Failure notification
Event deduplication
State machine update
User notification
Push notification
SMS (recipient)
Email receipt
In-app status
Prael conversation update
5 Feedback loop & model improvement
Outcome logging
Actual delivery time
Final success/fail
Error codes
Retry count
Total cost (actual)
Model retraining
Daily batch training
Feature drift detection
A/B test new model
Champion/challenger
Rollback if worse
Retrain cycle: daily
Partner feedback
Score adjustment
Blacklist (if SLA breach)
Volume rebalancing
Contract renegotiation
New partner evaluation
Analytics dashboard
Corridor performance
Partner comparison
Cost savings report
Speed improvements
Anomaly alerts
Cost savings: $2.3M/yr