Trust Bank resolves 52 percent of customer queries with AI using Amazon Bedrock
Discover how Trust Bank built a generative AI chatbot on Amazon Bedrock to resolve over 52 percent of customer queries without human intervention.
Benefits
- AI resolution rate monthly
- 52%+
- resolution for top contact types
- 80%+
Overview
Trust Bank, a digitally-native bank in Singapore backed by a unique partnership between Standard Chartered and FairPrice Group, grew to over one million customers in just over two years. It has also become one of the world’s fastest-growing digital banks and the fourth-largest retail bank in Singapore by customer numbers. With customer service requests scaling alongside that growth, the bank needed a solution that could deliver consistent, high-quality responses at scale without compromising the experience that fueled its growth. By building a generative AI chatbot on Amazon Bedrock, Trust Bank now resolves more than 52 percent of all customer queries without human intervention. The solution operates within a rigorous AI risk and governance framework that satisfies Monetary Authority of Singapore (MAS) regulatory requirements.
About Trust Bank
Trust Bank is Singapore's first digitally-native bank, backed by a unique partnership between Standard Chartered and FairPrice Group. With over one million customers, it is Singapore's fourth-largest retail bank by customer numbers. Regulated by the Monetary Authority of Singapore, Trust Bank has been named Singapore’s Best Digital Bank for Consumers by Euromoney.
Challenge | Sustaining award-winning service at scale
Trust Bank was launched in 2022 as a digitally-native bank backed by Standard Chartered and FairPrice Group. In just over two years, the bank surpassed one million customers, was rated the number one Finance App on the Singapore Apple App Store, and named Singapore’s Best Digital Bank for Consumers by Euromoney in 2025 and 2026. More than 70 percent of new sign-ups came through existing customer referrals, a signal that service quality directly fueled growth. But that same pace of expansion strained the bank's customer service operations.
Approximately 32 percent of the bank's chat volume consisted of simple generic enquiries handled by human agents. Each interaction averaged 15 minutes of handling time, making up a significant portion of the customer service team’s time. Human variability also introduced inconsistency despite high training standards, and the cost of a human-led model could not scale sustainably alongside the bank's growth trajectory.
As a digital bank with no physical branches, Trust Bank recognized that customer service quality is inseparable from customer experience. Trust Bank needed a solution that could resolve routine enquiries with consistent, accurate responses while maintaining its award-winning customer experience and remaining compliant with MAS regulations on data privacy, model governance, and responsible AI.
Solution | Building a regulated generative AI chatbot on Amazon Bedrock
Trust Bank had built its entire technology stack on AWS from day one, so extending into generative AI on the same infrastructure was a natural step. The team selected Amazon Bedrock as the backbone for its AI chatbot because it provides access to multiple foundation models, which the bank used to evaluate and choose the best-fit model. "Our golden rule is to choose the model that fits our purpose," said Stuart Loxton, Head of AI and Data Platform at Trust Bank. Bedrock's architecture, including AWS PrivateLink connectivity and native guardrails, made it the right fit for a MAS-regulated environment where customer data must never leave the bank's AWS environment or be shared with model providers.
The architecture routes customer messages through an API Gateway to a containerized chatbot AI agent running on Amazon EKS. The agent pre-fetches key customer profile and product information before invoking Bedrock foundation models for response generation. To ground answers in verified information, the team implemented Retrieval Augmented Generation (RAG) using Amazon Aurora as a vector database, storing embeddings generated through Bedrock.
Trust Bank also implemented a defense-in-depth approach using Amazon Bedrock Guardrails: denied topics, content filters, prompt injection protection, PII and PCI redaction. Beyond technical safeguards, the bank established an AI Risk Framework with gated approvals across four stages (Design, Proof of Concept, Beta, and Production). Each gate requires a customer journey and architecture review, foundation model assessment, responsible AI risk and controls assessment, security threat modeling, deployment, and testing. A Responsible AI (RAI) Council actively reviews and approves each stage before the solution advances to production.
The team validates model quality through automated testing, with thousands of test samples evaluated using an LLM-as-a-judge approach, combining human evaluation for ground truth, reference-based regression testing, and reference-free continuous monitoring of live outputs.
Outcome | Resolving over half of all customer queries with AI
Between February and September 2026, Trust Bank's generative AI chatbot achieved a monthly AI resolution rate exceeding 52 percent, meaning more than half of all customer enquiries reached a satisfactory conclusion without human intervention. For the bank's top contact type driver, the resolution rate exceeded 80 percent. An LLM-as-a-Judge flagged live conversations, and these were added to the quality dataset so each release was tested against real usage. This feedback loop was the main driver of better response quality and resolution rates over time. See Figure 1 below.
Each AI-powered chat interaction makes the marginal cost of serving a growing customer base essentially negligible. The bank can now scale to meet demand without proportionally increasing headcount, resolving the cost sustainability challenge that originally motivated the initiative. As a result, the bank’s customer service agents can focus on more complex customer enquiries. In addition, several agents have been retrained and redeployed into the bank’s AI operations team.
Trust Bank is now expanding its AI capabilities across the organization. Planned initiatives include a virtual AI relationship manager, code assistant tools for software development lifecycle enhancement, hyper-personalization engines, an internal knowledge base, and customer transaction search. Each initiative will follow the same gated governance framework that delivered the chatbot safely into production.
Here are three key lessons from Trust Bank's journey:
- Risk-focused governance is the enabler of production AI. Trust Bank's gated framework (Design to POC to Beta to Production) with executive oversight gave the risk committee confidence to approve deployment.
- Model choice matters. By selecting the best-fit foundation model for its specific use case rather than defaulting to one provider, Trust Bank optimized across speed, cost, quality, and security simultaneously.
- Automate testing before you automate customers. Trust Bank built an LLM-as-a-Judge evaluation framework using anonymized production data to validate quality before scaling AI responses to the full customer base. That investment in automated quality assurance is what allowed them to scale from 10 percent to 100 percent of traffic with confidence.
As a next step in serving customers, Trust Bank launched Trust AI Ask, a market-first in Singapore that lets customers search and understand their card spending through natural conversation. Rather than navigating menus or scrolling through transaction history, customers ask how much they spent on dining last month, at a specific merchant, or while travelling, and obtain totals and summaries in seconds. The feature was shaped by the insight that around 80 percent of active customers review their spending every month. Trust AI Ask uses only the individual customer's own data, with safeguards protecting personal information. Card spending is the starting point, with plans to extend to accounts, transfers, and rewards as the experience evolves toward personalized financial insights.
Figure 1: Trust Bank diagram
Amazon Bedrock gave us the flexibility to select the right models for different use cases, with governance capabilities built within a single managed service. As a result, more than 52% of customer enquiries are now resolved by AI, enabling us to scale support efficiently as our business grows.
Rajay Rai
Chief Information and Operating Officer, Trust BankAWS services used
More customer stories
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages