TELISIK/Performance

Detection Performance

Outcomes against the Proof-of-Value target bands · last 90 days.

AI-generated insightQwen-Max · trained on ALTO post-transaction history· Last 30 days vs PoV targetsConfidence 96%Configurable

False Positive Rate sits at 62% — just inside the 45–64% PoV band, down from a 91% baseline. True Positive Rate is 78%, comfortably inside the 77–87% band. Rule update cycle is now hours (vs weeks baseline), with full audit trail. Two drifting rules (R-014, R-118) have AI-drafted retunes pending human approval — adopting both projects FPR another ~6 pp lower.

Sapa drafts and recommends — you decide.

False Positive Rate

tracking to target

62%current

Baseline

91%

Target band

45–64%

lower is better

True Positive Rate (fraud catch)

in target band

78%current

Baseline

67%

Target band

77–87%

higher is better

Rule Update Cycle

weeks → hours

hourscurrent

Baseline

weeks

Target band

internal SLA

time to deploy a vetted rule change

Targets validated in adjacent FSI deployments; ALTO-specific figures confirmed during the Proof of Value.

FPR & TPR over 90 days

shaded = target band

Confusion Matrix (last 30 days)

True Positive

412

correctly caught fraud

False Positive

678

good tx wrongly flagged

False Negative

missed fraud

True Negative

18,420

correctly approved

Precision 37.8%Recall 81.1%

Rule Performance Leaderboard

Best 5 by precision

R-112	ATM Skimmer Hotspot ATM	62.1%
R-088	Dormant Account Reactivation QRIS MPM	55.0%
R-130	Beneficiary Fan-In QRIS CPM	47.3%
R-021	ATM Cash-Out Burst ATM	41.2%
R-101	Issuer-Acquirer Geo Velocity Debit POS/NPG	33.0%

Worst 5 by precision

R-014	QRIS High-Velocity Merchant QRIS MPM	9.1%
R-118	Refund Abuse Pattern Debit POS/NPG	18.5%
R-072	Cross-Border QRIS Mismatch QRIS CPM	22.4%
R-034	Debit POS Card-Not-Present Anomaly Debit POS/NPG	28.7%
R-101	Issuer-Acquirer Geo Velocity Debit POS/NPG	33.0%

Drift Detection

Rule Evaluator + Closed-Loop agents

Auto-flagged when a metric moves > 2σ from its rolling baseline.

R-014 · QRIS High-Velocity Merchant

FPR +3.1σ over last 14d (rising)

R-118 · Refund Abuse Pattern

TPR -2.4σ over last 14d (falling)

Analyst Throughput

Before TELISIK

cases / analyst / day

→

After TELISIK

cases / analyst / day

+100%

Cases closed per analyst per day · same analyst cohort · 30-day rolling average.