Malaysia's Ryt Bank Processes 80,000 Monthly Transactions Through Chat
Ryt Bank in Malaysia has replaced its mobile app's traditional menu structure with a conversational AI system that executes banking transactions directly. The system processes roughly 80,000 transactions per month across more than 50,000 users.
The bank deployed the regulator-approved assistant last summer. Customers now transfer funds, pay bills and check balances by typing natural-language requests instead of navigating multiple screens.
How the System Works
The AI interprets each request, prepares the transaction and displays it for customer approval before executing it. This confirmation step serves as a safeguard against unintended actions and helps the bank meet regulatory requirements.
The workflow operates in sequence: the system screens incoming messages for suspicious content, determines the user's intent, prepares the transaction details and waits for approval. Only then does it send instructions to the bank's payment infrastructure.
Ryt Bank built the system internally rather than adapting a commercial AI model. The bank developed its own large language model called ILMU through YTL AI Labs.
Language was a practical driver. Conversations in Malaysia often blend English, Malay and Chinese in single sentences, which commercial models struggle to handle without performance loss. A custom model trained on Malaysian conversational patterns solved this problem.
Regulatory control was equally important. Building the system in-house allowed the bank to maintain transparency over training data, updates and model behavior - something regulators demand for financial systems.
Testing showed the model recorded hallucination rates below 1.5 percent overall and less than 0.5 percent in high-risk financial workflows, according to the research paper describing the system.
Why Other Banks Haven't Followed
Banks of America and Capital One operate AI assistants, but these tools provide information or customer support rather than executing transactions directly. The difference matters for regulation and risk.
Large financial institutions operate across multiple jurisdictions, each with its own regulatory requirements for automated systems. Adding conversational AI that initiates payments would require additional approvals and risk frameworks that many banks consider too complex to justify.
Security presents another obstacle. Traditional banking apps use fixed menus and structured workflows. Chat interfaces accept open-ended language, creating new attack surfaces. Researchers have documented risks including prompt injection attacks, where malicious inputs attempt to manipulate how AI interprets instructions, and voice impersonation.
Operational risk remains the primary constraint. A PYMNTS survey found that 52% of CFOs would accept AI suggestions on liquidity and payment timing, but only 23% would allow it to coordinate finance workflows. The gap reflects how much skepticism persists about giving AI control over money movement.
Ryt Bank mitigates some risks through transaction confirmation and request screening, but broader questions persist: how do financial institutions verify not only who a customer is but also whether their conversational request reflects their actual intent?
The infrastructure supporting Ryt AI runs on Alibaba Cloud. The bank built its core platform in approximately six months. Customers who use the chat features tend to remain more active than those relying solely on traditional digital channels, according to the bank.
Your membership also unlocks:
