GX Bank integrates AI to scale fraud detection, risk management and business intelligence

Malaysia’s first licensed digital bank has built a number of in-house AI tools to manage its burgeoning transaction and loan volumes. Caroline Chong, Head of Data at GX Bank, discusses the FrAIdy and TrAIdy fraud detection frameworks, GuardPlus document forensics tool, and BI Bytes self-service analytics chatbot.

GXBank, Malaysia’s first digital bank license, received approval from Bank Negara Malaysia to commence operations in September 2023, becoming the first of five Malaysian digital bank license applicants. Since then, the number of customers has grown to more than 1.4 million, with more than 220 million transactions reported in 2025. We are also accelerating our business banking efforts, with approximately RM25 million (approximately $5.6 million) being disbursed to micro, small and medium enterprises (MSMEs) through GX FlexiLoan.

Growth has increased the volume and complexity of risk work. The bank has built three AI-enabled tools in-house. FrAIdy and TrAIdy for assessment of anti-fraud and anti-money laundering (AML) cases, GuardPlus for document forensics in digital lending, and BI Bytes for managed self-service business intelligence.

GXBank’s head of data, Caroline Chong, said these tools are designed to reduce operational costs and allow analysts and business teams to focus on higher-value work. The bank is positioning AI as an operational support layer, rather than a replacement for human judgment in regulated decision-making.

Growth increases risk and data workloads

As GXBank has grown to serve more than 1.4 million customers and process more than 450 million transactions, the volume and complexity of its risk operations has increased significantly. In traditional models, as trading volume and number of customers increase, analyst headcount must also increase proportionately.

GXBank has developed FrAIdy and TrAIdy as dual-engine generated AI frameworks for risk operations. The system uses transactional and behavioral data to create risk narratives and recommendations for analysts. It does not make the final decision. Human judges will still be responsible for particularly complex or high-risk cases.

FrAIdy and TrAIdy reduce case evaluation time

FrAIdy and TrAIdy automate the most time-consuming part of case processing: gathering information from multiple systems and converting it into a structured assessment. Chong said the tool has reduced case processing time from 15-20 minutes to 1-3 minutes while maintaining up to 95% accuracy in identifying high-risk cases and clearing low-risk alerts.

The bank scaled to 1 million customers without increasing its risk operations team. We also saved approximately 16,000 labor hours annually by reducing manual data capture.

“By automating routine tasks, we are able to handle higher case volumes and free up our analysts to focus on more complex problems,” said Chong.

The AI ​​model is based on real transaction data from GXBank’s data warehouse and is supported by pre-configured prompts and role-based access controls. Version 2.0 introduced more customer-centric evaluation logic and cost optimization, while the next phase will add batch processing and more advanced predictive analytics.

GuardPlus strengthens document checks in digital lending

In April 2025, the bank identified a sophisticated fraud case involving a falsified profit and loss statement within a digital lending application. This event necessitated a strategic realignment of credit operations, prompting banks to make temporary adjustments to lending limits while actively developing and implementing more robust security controls.

GuardPlus was built as an internal forensic layer to detect falsified P&L statements before they go through the lending process. Rather than focusing solely on the financial information displayed in the document, the tool analyzes file structure, metadata, trailer sequences, embedded objects, and content characteristics. This allows banks to test not only whether the numbers are valid, but also whether the documents have been falsified.

The system covers EPF statements, 98% of corporate bank statements and 94% of personal bank statements submitted to banks. The checks process on average in less than 0.1 seconds and achieved 100% accuracy, precision, and recall in stress tests using known fraudulent documents and forged test files created by our cybersecurity team.

This tool is integrated into the credit decision workflow as a real-time control. Its role is to prevent documents that show signs of tampering while maintaining direct processing for genuine applicants. GXBank is also developing a hybrid generative AI model to allow the system to more easily adapt to new bank statement formats and reduce manual rule maintenance.

BI Bytes gives business teams controlled access to data

BI Bytes was developed to relieve pressure on business intelligence (BI) teams who respond to more than 100 ad hoc requests a year, ranging from simple extractions to more complex analyses. This created delays for business users and forced analysts to focus on repetitive reports rather than high-value analysis.

This tool is a generative AI-powered chatbot developed in-house. This allows users in departments such as retail, marketing, finance, product, and operations to ask questions in plain English. The system translates these questions into Snowflake SQL, generates summaries of insights, and supports data visualization through a managed interface.

It achieved a 94% SQL accuracy rate during testing and is designed to absorb 30% of ad-hoc analysis requests across major departments. Chong predicted cost savings of RM800,000 (approximately $182,000) over three years by gaining back analyst time and reducing reliance on manual data acquisition.

“BI Bytes gives our team more self-directed access to business-critical data,” said Chong.

Rather than relying on third-party tools, GXBank created its own semantic layer that incorporates validated banking logic, table definitions, and approved query structures. This design aims to reduce the risk of hallucinations and ensure that self-service analytics are controlled within a regulated environment.

“Although AI is a tool to support our team, we ensure that the final decision is made by our analysts, especially when there are higher stakes or potential risks,” Chong said.

The next phase will focus on scale and adaptability.

GXBank’s next phase will focus on improving the three systems and expanding their use across the bank. FrAIdy and TrAIdy are expected to move towards more advanced batch processing and predictive analytics. GuardPlus is developed as a hybrid generative AI solution that can adapt to new document formats. BI Bytes will be expanded to more data domains such as MSME, mobile events, and risk.

As GXBank continues to grow, the ability of these tools to maintain accuracy, control, and auditability will be a key test for the still-infant digital bank.