Project Name
Real-Time Fraud Screening for 100% of Card & ATM Transactions with Informatica
The client is a retail bank that screens card and ATM transactions using a fraud scoring engine. The engine assesses each transaction against recent customer behaviour, but it was fed by a nightly batch extract. The bank needed every transaction scored as it happened, with full accountability and no changes to its core banking system.
- Next-Day Fraud Detection: Transactions reached the fraud team through a nightly extract. Fraudulent activity could continue for hours before review.
- Incomplete Notifications: Core banking notifications carried only a reference. The scoring engine required a complete document with account holders and risk attributes.
- Mixed Notification Queue: Card, ATM, transfer, and account messages shared one queue. A fixed-schema listener would drop non-matching messages without error.
- Release-Bound Risk Signals: Every new risk attribute required an interface mapping change. The interface became a bottleneck for the fraud team.
- Outage and Duplicate Risks: Engine outages either needed manual replays or looped indefinitely. Redelivered messages could also trigger duplicate scores and cases.
- Invisible Lost Transactions: Transactions lost between systems left no trace. Nobody could review what never arrived.
Ksolves built an event-driven integration on Informatica Cloud Application Integration between the core banking system and the fraud scoring engine. It filters, enriches, and scores every in-scope transaction in real time, tracks each one in an audit table, and writes nothing back to the core system.
- Real-Time Listener with Early Filtering: A JMS listener accepts any XML message and passes on only card and ATM debits. Other types are ignored before any database or API call.
- On-the-Fly Enrichment: Each notification is completed from a read-only transaction view. It is converted into the nested JSON the engine expects, including joint account holders.
- Pass-Through Risk Attributes: Custom risk attributes travel as one opaque field, copied verbatim into the payload. New signals reach the engine with zero interface changes.
- Three-Queue Retry Design: Failed engine calls move to a retry queue with an attempt counter. After the agreed limit, messages park on a failed queue for replay.
- Audit-First Duplicate Protection: Each message is checked against the audit table before processing. Completed transactions are dropped, and interrupted ones resume safely.
- Business-Aware Exceptions: Missing transactions are marked NOT_FOUND and raised on an alert queue. HIGH-risk verdicts go straight to the fraud team's case tool.
Technology Stack
| Category | Technology |
|---|---|
| Integration | Informatica Cloud Application Integration |
| Integration | JMS Message Broker |
| Database | Relational DB (View + Audit Table) |
| Platform | Fraud Scoring Engine (REST API) |
| Infrastructure | Informatica Secure Agent |
- Real-Time Fraud Screening - Before: transactions were reviewed the next morning. After: every card and ATM transaction is scored within moments of posting.
- 1 Audit Record per Transaction - Before: lost transactions left no trace. After: every in-scope transaction ends with exactly 1 audit record and a final status.
- 0 Transactions Lost to Outages - Before: outages required manual replays or looped endlessly. After: a 3-queue retry design delays verdicts but never drops them.
- 0 Interface Changes for New Signals - Before: each new risk attribute needed a release. After: new attributes reach the engine with no interface change.
- 8 Failure Paths Tested at Sign-Off - Before: failures went unnoticed. After: 8 deliberate failure tests show that duplicates, outages, restarts, and missing data are all handled.
“We now flag suspicious card activity while it is still happening, and we can account for every transaction from posting to verdict.”
– Head of Fraud Operations, retail bank.
Ksolves replaced overnight fraud review with real-time screening on Informatica Cloud Application Integration, so every card and ATM transaction is scored within moments of posting. Each transaction is tracked from posting to verdict through a single audit record. Outages, restarts, and duplicates are handled without losing a single transaction, and new risk signals reach the engine without interface releases. Built on a reusable request-and-verdict contract, the framework can connect to any scoring engine as the bank’s fraud strategy evolves.
Is your fraud team still reviewing yesterday’s transactions?
