TELISIK
powered by Alibaba Cloud
TELISIK/Rules

Rule Management

AUGMENT — rule lifecycle. AI recommends, a human approves every change.

New members start with Bank Indonesia-recommended baseline rules pre-loaded.

AI-generated insightQwen-Max · trained on ALTO post-transaction history· Network rule lifecycleConfidence 94%Configurable

Two active rules are drifting this week: R-014 (FPR +3.1σ) and R-118 (TPR −2.4σ). The Rule Evaluator has drafted retunes — every parameter is configurable and nothing deploys without a human approver. Approving R-014 v2 alone would cut network-wide false positives by ~9%.

Sapa drafts and recommends — you decide.
Rule catalog (12)
IDNameStatusPrecisionTriggers (30d)
R-014QRIS High-Velocity Merchant
QRIS MPM
Active9.1%1,802
R-021ATM Cash-Out Burst
ATM
Active41.2%412
R-034Debit POS Card-Not-Present Anomaly
Debit POS/NPG
Active28.7%612
R-072Cross-Border QRIS Mismatch
QRIS CPM
Active22.4%198
R-088Dormant Account Reactivation
QRIS MPM
Active55.0%88
R-101Issuer-Acquirer Geo Velocity
Debit POS/NPG
Active33.0%274
R-110Merchant KYC Similarity Cluster
QRIS MPM
Draft0.0%0
R-112ATM Skimmer Hotspot
ATM
Active62.1%47
R-118Refund Abuse Pattern
Debit POS/NPG
Active18.5%320
R-130Beneficiary Fan-In
QRIS CPM
Active47.3%105
R-141Retired Legacy ATM Velocity
ATM
Retired4.0%0
R-152QRIS Cross-Member Pattern (β)
QRIS MPM
Draft0.0%0
R-014ActiveOwner: ALTONeeds attention

QRIS High-Velocity Merchant

QRIS MPM

Rule Definition
IFchannel = QRIS_MPM
ANDcount(merchant_id, window=10min) > 25
ANDavg(amount) < 300,000
THENflag = HIGH
Show raw DSL (audit view)
IF channel = QRIS_MPM AND count(merchant_id, window=10min) > 25 AND avg(amount) < 300,000 THEN flag = HIGH
Live performance (last 30 days)
Triggers
1,802
Confirmed fraud
127
False positives
1,294
Precision
9.1%
AI recommendationQwen-Max · Rule Evaluator agentConfidence 92%
Tighten R-014 velocity threshold 25 → 35 and gate by merchant_age
Generated for R-014 — QRIS High-Velocity Merchant · derived from post-transaction history (alerts, analyst dispositions, confirmed-fraud outcomes)
  1. 1Of the 4,217 R-014 triggers in the last 90 days, 91% were dispositioned as false positives by analysts (3,837 of 4,217).
  2. 294% of false positives came from 3 merchant categories: top-up agents (MCC 6051), parking lots (MCC 7523) and street-food vendors (MCC 5812) — all of which legitimately exceed 25 small payments / 10 min.
  3. 3Confirmed-fraud triggers cluster at 38+ payments / 10 min and merchant_age < 90 days — raising the threshold to 35 and adding a merchant_age gate preserves catch.
  4. 4Replaying the proposed rule across 90d of real traffic suppresses 41% of false positives while only dropping 3% of confirmed-fraud catches (none of which were unique to this rule — all were also flagged by R-072).
Proposed parameters — fully configurable
Velocity threshold
25 txns / 10 min35 txns / 10 min
Merchant age gate
none days< 90 days
Min amount
IDR10,000 IDR
Flag severity
HIGHHIGH
Projected impact (from historical backtest)
False positives
-41%
Confirmed catch
-3%
Net precision
+17.9 pp

AI-generated from the last 90 days of post-transaction outcomes. Every parameter above is configurable; no rule deploys automatically — a human approver must accept the change. All actions are logged to ActionTrail.

Prototype — mock data