While many companies struggle to find practical applications for AI, DBS Bank is already running over 1,500 AI and machine learning models across more than 370 use cases, from credit risk modelling to early detection of portfolio vulnerabilities. In 2023, its AI initiatives generated $370 million in cost savings and value creation, a figure the bank expects to exceed $1 billion this year.
So, how has DBS managed not only to keep pace with AI's rapid development but also to extract tangible business value from the technology consistently? The answer lies in taking a platform-first and open architecture approach that enables AI to be industrialised at scale.
Since data quality and accessibility are critical to scaling AI, DBS built an internal self-service platform called ADA (or Advancing DBS with AI), which serves as a single source of truth for data governance, discoverability, quality and security.
It also developed ALAN, an AI protocol platform that acts as both a library for employees to reference when building new projects and a repository to track published models and best practices. With ADA and ALAN, DBS teams are equipped to develop and deploy compliant AI models quickly, driving improvements in operations and decision-making.
Hybrid IT architecture
To further strengthen its capabilities, DBS has adopted a hybrid IT architecture that combines on-premises data centres with cloud-based services. This approach allows the bank to scale AI and IT services quickly, lower infrastructure costs by using cloud solutions where suitable and enhance overall business resilience.
See also: Safeguarding elections in the age of artificial intelligence
Moreover, the IT architecture emphasises openness and interoperability, which offers the flexibility to use fit-for-purpose solutions rather than being tied to a single tech vendor. This flexibility is crucial for scaling generative AI across the bank.
"Since our IT architecture is vendor-agnostic, we can deploy the most suitable large language model (LLM) from any provider for a specific use case, or change models quickly," says Nimish Panchmatia, DBS Bank's chief data and transformation officer, to The Edge Singapore.
He likened the architecture to a multi-plug, enabling the bank to seamlessly connect to a range of AI models or LLMs while maintaining strict control over which ones are used. This design also helps manage and optimise AI costs, an increasing concern as organisations scale their use of generative AI. With telemetry, the bank can monitor model performance in real time and switch to more cost-effective alternatives as they become available, ensuring fiscal discipline in its AI deployments.
See also: Alibaba rolls out latest flagship AI model in post-DeepSeek race
DBS's selection of Google Cloud as its preferred cloud AI services provider underscores the strategic value of its open ecosystem. By integrating ADA and ALAN with Google Cloud's Vertex AI Model Garden, DBS's AI engineers can access over 1,500 in-house machine learning models, along with more than 200 first-party, third-party, and open-source models from providers like Google, Anthropic and Meta.
This flexibility allows DBS teams to fine-tune and deploy their chosen models to endpoints through standardised workflows, while continuously evaluating and switching models via managed APIs (or model-as-a-service) to optimise AI's performance. Additionally, Vertex AI's Provisioned Throughput for the Gemini and Claude models provides DBS with assurance that it can scale its generative AI applications reliably and cost-effectively, even during periods of high demand.
Generative AI at work
According to Panchmatia, DBS adopts a structured approach to developing generative AI use cases to maximise business impact. These use cases are divided into two categories: vertical capabilities, where generative AI serves as a co-pilot for specific roles, such as relationship managers, software developers, and operations teams, and horizontal capabilities, which address a wide range of employee needs across the organisation.
The Customer Service Officer (CSO) Assistant is one of the vertical use cases for generative AI developed by the bank's AI engineers. Powered by Google Cloud's Customer Engagement Suite with Google AI, the CSO Assistant is tailored to local nuances and language, incorporating voice telephony and speech recognition.
It offers real-time transcription and immediate access to the bank's knowledge base, reducing the time CSOs need to provide solutions by 20%. Additionally, it supports post-call documentation with instant summaries and pre-filled service requests. The CSO Assistant is currently used by 500 CSOs in Singapore to improve the quality of service for over 250,000 customers who contact the bank each month.
Meanwhile, DBS-GPT is a horizontal generative AI application currently available to 90% of the bank's workforce. Built on Google's Gemini models and other LLMs, DBS-GPT serves as a multipurpose AI assistant for content generation and writing tasks within a secure environment.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
Using Gemini's multimodal capabilities and long context window, DBS-GPT can process a large volume of information in multiple formats, including text, images, audio and video. This allows employees to quickly extract specific details or insights from internal documents and knowledge repositories, significantly accelerating tasks that were once labour-intensive and time-consuming.
Preventing rogue AI
Even with access to vast amounts of high-quality data, AI models - particularly large language models - can still hallucinate, generating incorrect, misleading or entirely fabricated information. Aware of this risk, DBS has introduced safeguards to minimise hallucinations as it expands the use of generative AI across the organisation.
"We have a robust framework around responsible AI and every AI use case is assessed by the Responsible Data Use (RDU) Committee [before it goes live and into production]," says Panchmatia.
DBS enforces robust data governance through its PURE (purposeful, unsurprising, respectful and explainable) framework and aligns its approach with the Monetary Authority of Singapore's FEAT principles, which prioritise fairness, ethics, accountability and transparency. Oversight is further reinforced by the bank's RDU Committee, a cross-functional team of senior leaders tasked with reviewing data usage, identifying inherent risks and ensuring appropriate safeguards are in place across its AI applications.
He adds that keeping a "human in the loop" will remain critical as generative AI adoption grows. Reflecting this approach, DBS's generative AI use cases are currently designed to support employees in working more productively and effectively, rather than replacing them.
"Even as [generative] AI simplifies tasks and automates some decision-making, at the end of the day, a human will still be there [at the final step to deliver the service]," he says. "The main focus of AI should be on helping us provide better advice and services to our clients, to make them successful."
Strategic AI integration
It takes more than simply deploying an LLM into a business process to find tangible business value with AI at scale. "For AI to be embedded into the organisation, you'll need to re-engineer processes. Changing a process [is complex and] requires careful consideration as it could have knock-on effects on other processes. For us, we determine the desired outcomes before deciding the shifts needed in the processes and where to embed AI in the workflows," says Panchmatia.
He continues: "By doing so, we can get employees to see the need for change and bring them along the transformation journey. You can have the best technology or ecosystem in the world, but if your employees are not with you or you don't have their buy-in, you'll fail in your transformation journey."
Recognising the vast potential of AI, DBS plans to keep exploring new ways to harness the technology, including machine learning, deep learning and AI agents that can perform tasks on behalf of users, to enhance operational efficiency.
Panchmatia adds that the bank is also exploring ways to use AI to deliver more services in Asian markets where it has a presence but a limited physical footprint, much like it did with its mobile-first digibank platform, which extended its reach in Indonesia and India.
