Blab AI
GXBank Deploys AI-Powered Tools to Manage Rapid Growth and Fraud Risks
GXBank, Malaysia’s first licensed digital bank, has surged past 1.4 million customers and handled more than 450 million transactions in 2025. To keep pace with this expansion—and the accompanying spike in risk‑management demands—the bank has built three internal AI‑enabled systems: FrAIdy and TrAIdy for fraud and anti‑money‑laundering (AML) case assessment, GuardPlus for document forensics in digital lending, and BI Bytes for governed self‑service business intelligence.
The bank’s rapid growth has amplified both the volume and the complexity of its risk operations. In a traditional model, higher transaction and customer counts would have required a proportional increase in analyst headcount. Instead, GXBank’s FrAIdy and TrAIdy employ a dual‑engine generative‑AI framework that draws on transaction and behavioural data to produce risk narratives and recommendations for human reviewers. The tools do not make final decisions; analysts retain accountability, especially for complex or higher‑risk cases. According to Caroline Chong, GXBank’s Head of Data, the systems were designed to lower operating costs and free analysts and business teams to focus on higher‑value work.
FrAIdy and TrAIdy have cut case‑processing time from 15–20 minutes to just 1–3 minutes while maintaining up to 95 % accuracy in flagging high‑risk cases and clearing low‑risk alerts. The bank has scaled to one million customers without adding headcount to its risk‑operations team, saving roughly 16,000 working hours annually by reducing manual data retrieval. The AI model is grounded in factual transaction data from GXBank’s data warehouse, supported by pre‑configured prompts and role‑based access controls. Version 2.0 introduced more customer‑centric assessment logic and cost optimisation, and the next phase is expected to add batch processing and more advanced predictive analytics.
In April 2025, GXBank uncovered a sophisticated fraud incident involving tampered income statements in its digital lending applications. The event prompted a temporary adjustment to lending limits while the bank developed GuardPlus, an internal forensic layer that detects manipulated income statements before they move through the lending process. GuardPlus analyses the file’s structure, metadata, trailer sequences, embedded objects and content characteristics, rather than just the financial information displayed. The system covers EPF statements, 98 % of company bank statements and 94 % of individual bank statements submitted to the bank. It processes checks in less than 0.1 seconds on average and achieved 100 % accuracy, precision and recall in stress tests using known fraudulent documents and forged test files prepared by the cybersecurity team. GuardPlus is integrated into the credit‑decisioning workflow as a real‑time control, preserving straight‑through processing for genuine applicants while stopping documents that show signs of tampering. GXBank is also developing a hybrid generative‑AI model so that the system can adapt more easily to new bank‑statement formats and reduce manual rule maintenance.
BI Bytes was created to ease pressure on a business‑intelligence team that fielded more than 100 ad‑hoc requests a year. The generative‑AI‑powered chatbot allows users in retail, marketing, finance, product and operations to ask questions in plain English. The system translates those questions into Snowflake SQL, generates insight summaries and supports data visualisation through a governed interface. It achieved a 94 % SQL accuracy rate during testing and is designed to absorb 30 % of ad‑hoc analytics requests across key departments. Chong projected MYR 800,000 (approximately $182,000) in cost savings over three years by reclaiming analyst time and reducing dependence on manual data pulls. Rather than relying on third‑party tools, GXBank created a proprietary semantic layer that embeds verified banking logic, table definitions and approved query structures. This design is intended to reduce hallucination risk and ensure that self‑service analytics remains controlled within a regulated environment.
GXBank’s next phase will focus on improving the three systems and extending their use across the bank. FrAIdy and TrAIdy are expected to move into more advanced batch processing and predictive analytics. GuardPlus is being developed into a hybrid generative‑AI solution that can adapt to new document formats. BI Bytes will be expanded into additional data domains, including MSME, mobile events and risk. Whether these tools can maintain accuracy, control and auditability as GXBank continues to grow will be an important test for a digital bank that is still in its early years.
The bank’s approach illustrates how a regulated financial institution can use AI as an operational support layer rather than a substitute for human judgement. By automating routine tasks, GXBank has been able to handle a larger volume of cases, freeing analysts to focus on more complex issues while maintaining regulatory compliance and auditability.
The bank’s rapid growth has amplified both the volume and the complexity of its risk operations. In a traditional model, higher transaction and customer counts would have required a proportional increase in analyst headcount. Instead, GXBank’s FrAIdy and TrAIdy employ a dual‑engine generative‑AI framework that draws on transaction and behavioural data to produce risk narratives and recommendations for human reviewers. The tools do not make final decisions; analysts retain accountability, especially for complex or higher‑risk cases. According to Caroline Chong, GXBank’s Head of Data, the systems were designed to lower operating costs and free analysts and business teams to focus on higher‑value work.
FrAIdy and TrAIdy have cut case‑processing time from 15–20 minutes to just 1–3 minutes while maintaining up to 95 % accuracy in flagging high‑risk cases and clearing low‑risk alerts. The bank has scaled to one million customers without adding headcount to its risk‑operations team, saving roughly 16,000 working hours annually by reducing manual data retrieval. The AI model is grounded in factual transaction data from GXBank’s data warehouse, supported by pre‑configured prompts and role‑based access controls. Version 2.0 introduced more customer‑centric assessment logic and cost optimisation, and the next phase is expected to add batch processing and more advanced predictive analytics.
In April 2025, GXBank uncovered a sophisticated fraud incident involving tampered income statements in its digital lending applications. The event prompted a temporary adjustment to lending limits while the bank developed GuardPlus, an internal forensic layer that detects manipulated income statements before they move through the lending process. GuardPlus analyses the file’s structure, metadata, trailer sequences, embedded objects and content characteristics, rather than just the financial information displayed. The system covers EPF statements, 98 % of company bank statements and 94 % of individual bank statements submitted to the bank. It processes checks in less than 0.1 seconds on average and achieved 100 % accuracy, precision and recall in stress tests using known fraudulent documents and forged test files prepared by the cybersecurity team. GuardPlus is integrated into the credit‑decisioning workflow as a real‑time control, preserving straight‑through processing for genuine applicants while stopping documents that show signs of tampering. GXBank is also developing a hybrid generative‑AI model so that the system can adapt more easily to new bank‑statement formats and reduce manual rule maintenance.
BI Bytes was created to ease pressure on a business‑intelligence team that fielded more than 100 ad‑hoc requests a year. The generative‑AI‑powered chatbot allows users in retail, marketing, finance, product and operations to ask questions in plain English. The system translates those questions into Snowflake SQL, generates insight summaries and supports data visualisation through a governed interface. It achieved a 94 % SQL accuracy rate during testing and is designed to absorb 30 % of ad‑hoc analytics requests across key departments. Chong projected MYR 800,000 (approximately $182,000) in cost savings over three years by reclaiming analyst time and reducing dependence on manual data pulls. Rather than relying on third‑party tools, GXBank created a proprietary semantic layer that embeds verified banking logic, table definitions and approved query structures. This design is intended to reduce hallucination risk and ensure that self‑service analytics remains controlled within a regulated environment.
GXBank’s next phase will focus on improving the three systems and extending their use across the bank. FrAIdy and TrAIdy are expected to move into more advanced batch processing and predictive analytics. GuardPlus is being developed into a hybrid generative‑AI solution that can adapt to new document formats. BI Bytes will be expanded into additional data domains, including MSME, mobile events and risk. Whether these tools can maintain accuracy, control and auditability as GXBank continues to grow will be an important test for a digital bank that is still in its early years.
The bank’s approach illustrates how a regulated financial institution can use AI as an operational support layer rather than a substitute for human judgement. By automating routine tasks, GXBank has been able to handle a larger volume of cases, freeing analysts to focus on more complex issues while maintaining regulatory compliance and auditability.
