Generative AI has the potential to transform how organizations operate, but in highly regulated industries such as banking, innovation cannot come at the expense of accountability, transparency, or customer trust.
DBS set out to demonstrate how Generative AI can be deployed responsibly at scale in production environments where decisions affect customers, employees, and business outcomes.
Rather than treating GenAI as a standalone technology experiment, DBS designed a scalable GenAI CoPilot operating model embedded within three critical operational workflows: Consumer Banking Deposit Account Opening (CASA AOS), Self-Service Banking (SSB) Claims, and Wealth Management Corporate Actions. These workflows depend heavily on interpreting large volumes of unstructured information, including customer documents, transaction records, intervention reports, and technical SWIFT messages.
The objective was twofold:
Most importantly, DBS established a repeatable and scalable operating model that could be applied across different GenAI use cases and operational workflows within the bank. By embedding governance, human oversight, explainability, and risk controls into the design from the outset, DBS demonstrated that Responsible AI is not a barrier to innovation, but a foundation for sustainable and trustworthy transformation.
DBS recognised that deploying Generative AI in banking required more than technological capability. It required trust, accountability and strong governance. In workflows involving customer information, financial outcomes and client communications, innovation had to be balanced with accountability, transparency and human judgement.
Rather than introduce GenAI as a standalone tool, DBS designed a Responsible GenAI CoPilot operating model that embeds AI directly into existing workflows while keeping employees firmly in control. The model was deployed across three distinct operational areas: Consumer Banking Deposit Account Opening, Self-Service Banking Claims and Wealth Management Corporate Actions. Together, these use cases tested GenAI across diverse forms of unstructured information, from customer documents and transaction records to intervention reports and complex SWIFT messages.
Each implementation applied a common principle: AI augments people; it does not replace human accountability. In Account Opening, GenAI classifies and extracts information across up to 18 document types. In SSB Claims, it analyses transaction records and supporting evidence to generate recommendations. In Corporate Actions, it converts technical messages into clear, client-ready narratives. Every AI-generated output remains subject to 100% human validation before any decision or customer-impacting action is taken.
Responsibility was built into the solution through multiple safeguards, including maker-checker controls, audit trails, explainability and traceability, continuous model monitoring, manual fallback processes and kill-switch capabilities. These controls were reinforced through DBS’ Responsible Data Use and Responsible AI frameworks, creating clear accountability around how GenAI outputs are reviewed and used.
Deployment followed a phased approach involving experimentation, parallel runs, prompt refinement, golden-dataset evaluation and controlled production implementation. This allowed teams to identify edge cases, evaluate model behaviour against real-world scenarios and strengthen controls before broader adoption. Cross-functional collaboration among operations, technology and data teams ensured that the solutions were technically robust, operationally practical and responsibly governed.
Crucially, DBS treated adoption as a people transformation, not merely a technology rollout. Operations teams built familiarity with GenAI through parallel runs, output validation and iterative refinement, helping them understand both its capabilities and limitations. Positioning the technology as a CoPilot shifted behaviour from manually processing information towards critically reviewing AI-generated outputs, exercising judgement and managing exceptions.
By combining governance, technology and human expertise within one repeatable model, DBS demonstrated that Responsible AI can move beyond principles and become an operational reality across multiple banking domains.
The initiative demonstrated that Generative AI can be operationalised across critical banking workflows while preserving human accountability, transparency and control.
Across three distinct use cases, GenAI delivered strong and measurable performance when interpreting complex, unstructured information. The Account Opening solution achieved 99.55% document-classification accuracy and 91.79% entity-extraction accuracy across up to 18 document types. The SSB Claims CoPilot achieved 86.5% accuracy, while the Corporate Actions solution achieved 83% production accuracy and approximately 93% accuracy in controlled evaluation environments when transforming technical messages into client-ready narratives.
Just as importantly, all three implementations operated within a consistent governance model. AI outputs remained subject to 100% human validation, supported by maker-checker controls, audit trails, model monitoring and manual fallback processes. This demonstrated that responsible controls could be applied systematically across use cases with different data, risks and operational contexts.
The programme also changed how employees engage with GenAI. Operations teams moved from manually interpreting and consolidating information towards validating recommendations, applying judgement and managing exceptions. Through testing, parallel runs and feedback, employees developed greater confidence in using GenAI while remaining alert to its limitations and accountable for final outcomes.
Most importantly, DBS established a repeatable blueprint for responsible, human-centred GenAI adoption. By embedding governance, explainability, auditability and human oversight into the operating model, DBS showed that AI can support high-impact banking processes without weakening control or accountability. The result is not simply wider technology adoption, but a more thoughtful and responsible way for people and AI to work together.