Ryt Bank deployed the regulator-approved conversational AI system last summer, with the capability to execute core banking transactions. Customers use the bank’s Ryt AI assistant to transfer funds, pay bills and check balances through natural-language requests routed directly to the bank’s payment systems.
The assistant replaces the traditional mobile banking interface. Instead of browsing menus and multiple screens inside an app, customers interact with the bank through conversation. Ryt AI interprets the request, prepares the transaction and presents it for approval before funds move.
Conversational Workflow With Human Confirmation
Ryt AI converts customer conversations into banking actions through a structured process that interprets requests while maintaining regulatory safeguards.
Researchers describing the system say the workflow relies on several artificial intelligence components that work sequentially to evaluate requests and process transactions.
The system first screens incoming messages for suspicious or inappropriate prompts. It then determines the user’s intent, such as transferring funds or paying a bill. Once the request is understood, the system prepares the transaction and presents it to the customer for review.
Advertisement: Scroll to Continue
Customers must approve each transaction before it is executed. The system displays payment or transfer details and requires confirmation before sending instructions to the bank’s payment infrastructure. That step helps the bank meet regulatory requirements for transaction verification and reduces the risk of unintended actions.
Ryt Bank built the AI system internally rather than relying on an external model. The bank developed its own large language model, called ILMU, through YTL AI Labs, according to ComputerWeekly.
One reason was language. Conversations in Malaysia often combine English, Malay and Chinese in the same sentence, making it difficult to adapt commercial AI models without sacrificing performance. Building its own system allowed the bank to train the model specifically for Malaysian conversational patterns.
Regulation also influenced the decision. Internal development allows the bank to maintain control over training data, system updates and model behavior, giving regulators greater transparency into how the AI operates.
The bank evaluated ILMU using about 2,000 conversational scenarios, including multi-step interactions along with transaction requests. According to the research paper, the model recorded hallucination rates below 1.5 percent overall and less than 0.5 percent in high-risk financial workflows.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Infrastructure, Scale and Industry Constraints
The infrastructure supporting Ryt AI runs on Alibaba Cloud, which hosts the digital bank’s systems and computing environment. Ryt Bank launched its core banking platform in about six months using the cloud provider’s infrastructure and security framework, according to TechNode.
The system now serves more than 50,000 users and processes roughly 80,000 transactions per month, ComputerWeekly said. The bank says customers who use the chat-based features tend to remain more active than those who rely only on traditional digital channels.
Even so, the approach may be difficult for large financial institutions to replicate.
Many banks already operate AI assistants, including Bank of America’s Erica and Capital One’s Eno. But those tools generally provide information or customer support rather than executing financial transactions directly.
Large banks also operate across multiple regulatory jurisdictions. Each regulator requires clear oversight of automated financial systems, including audit trails and controls that stop unintended transactions. Deploying conversational artificial intelligence that can initiate payments would likely require additional approvals and risk frameworks.
Security is another concern. Traditional banking apps rely on structured workflows and fixed menus. Chatbot platforms accept open-ended language, which can create new opportunities for social engineering or manipulation.
Operational risk is one of the reasons more businesses haven’t given AI greater control over money movement, according to PYMNTS data. While 52% of CFOs surveyed for a recent PYMNTS report were agreeable to AI making suggestions on liquidity and payment timing, just 23% of all CFOs surveyed would allow it to coordinate finance workflows.
Researchers studying AI systems have identified risks such as prompt injection, in which malicious inputs attempt to influence how an AI interprets instructions, and voice impersonation attacks.
Ryt Bank addresses some of these risks by requiring confirmation before money moves and by screening incoming requests before processing them. Even so, broader questions remain about how financial institutions verify not only a customer’s identity but also the intent behind a request made through conversation.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
Add as Preferred Source
