Fraud Detection Assistant with ING
KickstartAI x ING
Through a collaborative effort, ING and KickstartAI are leveraging Generative AI and natural language understanding to deliver two key outcomes: an empathetic, consumer-facing triage system for rapid assessment of potentially fraudulent communications, designed to help customers recognize potential scam indicators earlier and strengthen collective resilience against social engineering fraud. This initiative supports fraud-prevention efforts through aggregated and privacy-conscious insights derived from user interactions that bridges the gap between vulnerable consumers and complex bank security protocols.
The Problem
The financial industry faces a complex challenge in staying ahead of evolving fraud tactics and sophisticated social engineering. While banks have robust backend security, a significant portion of consumers still struggle to recognize red flags when targeted directly.
In practice: A customer may receive a deceptive text message or a high-pressure phone call that bypasses traditional security filters by exploiting human psychology. Because scammers isolate victims in the critical moments before an irreversible transaction occurs, there is a "blind spot" where customers lack an easy, immediate way to verify if a situation is suspicious before they act.
Goals
The goal is to develop an empathetic conversational AI assistant (project name: Savvy Auntie) integrated directly into the ING banking app using ING's native Agent Development Kit (ADK) infrastructure. By providing real-time threat analysis, interactive guidance, and verified protection protocols within the existing client workflow, we aim to help customers identify and respond to potential social engineering attempts at an early stage.
Approach
In-App Triage & Context Verification: We begin by embedding conversational verification directly into the client app journey, enabling users to instantly check suspicious messages, links, or payment requests within their trusted banking environment.
ADK-Native Development: Using the Agent Development Kit (ADK), we build secure agent workflows aligned with ING security and privacy requirements, ensuring seamless interoperability with internal banking APIs.
Adaptive Security Integration: Statistically significant scam patterns and psychological triggers identified during interactions are structured to inform existing security filters, helping inform and improve existing fraud-detection capabilities over time.
Key features
Conversational Triage: Uses GenAI to guide users through suspicious scenarios with empathetic clarifying questions and verified steps for protection.
Multi-modal Threat Extraction: Employs image and screenshot analysis to helps assess potential phishing indicators, sender information and urgency cues commonly associated with scams.
Context-Aware PII Scrubbing: Uses an automated gateway to permanently separate and strip sensitive identifiers like names, IBANs, and home addresses before logging.
Expected Results
Operational Efficiency: Potential reduction in fraud-related support calls into automated digital verification, which may help improve operational efficiency and support fraud-prevention efforts.
Quality & Experience: Lowered barriers to security support, fostering higher cyber-resilience, reducing financial loss, and boosting customer confidence in digital banking.