Blogs

Attribute to: Mr. Chris Leung, Data & Technology Transformation Leader, Hong Kong, IBM Consulting

Artificial Intelligence (AI) has undoubtedly become a transformative force across various industries, with its impact now strongly being felt in the financial sector, including banking. This sector necessarily operates in highly regulated environments due to the critical role it plays in our society, as well as how coupled the stability of financial system is on banks. The value realization of generative AI – specifically in the banking sector – on revolutionizing customer experiences and operational efficiencies, require the proper set-up of risk management and governance structures to uphold trust towards data and AI technologies.

Today, almost 8 in 10 institutions (78%) from banking and financial industry are currently implementing generative AI with different use cases. Almost 60% of generative AI decision makers see the value of this cutting edge technology in risk management, compliance reporting, and client engagement.ⁱ

At the IBM Hong Kong Technology Forum – the Panel Discussion of “Scaling the Impact of Generative AI with Trusted Data & Governance," we sat down with industry experts from prominent banks, including The Bank of East Asia, Limited (BEA), Hang Seng Bank Limited and DBS Bank (Hong Kong) Limited, to share their insights on leveraging generative AI in the banking industry, amidst regulatory pressure, and provide valuable guidance for other organizations seeking to navigate this transformative landscape.

Personalized AI solutions for business integration and risk mitigation

Mr. Kenny Au, Acting Head of Operations Division at The Bank of East Asia, Limited (BEA), spoke at the forum leveraging his extensive experience in both digital transformation and business operations for different leading banks.

He recognized generative AI’s ability to enable hyper-personalization at scale that would fundamentally change how banks provide value to their customers. He also emphasized the value of AI technologies that will be unlocked to their fullest potential when tailored with specific set of business knowledge, and designed to be embedded in day-to-day business operations. Integrating AI models to existing business operations requires close collaboration with business units, well-defined change management planning and, most importantly,  re-thinking the future operating model with AI agents.

Kenny acknowledged the governance and risk management challenges associated with AI and stressed the need for robust risk mitigation strategies, addressing data security, operational risks, data privacy, and model risks.

Banks, including his own, are committed to defining the risk governance framework that is relevant to the latest technologies, and assembling experts across different departments to assess and review risks associated with AI models. Besides compliance requirements, the ethical use of AI technologies are also crucial considerations, including how banks ensure they apply models fairly to customers.

Kenny expressed his belief that investing in people is key even when the transformation scope is broad. At BEA, the data experts, key users and process experts work closely to understand what generative AI is and how it can be embedded into the processes, to ensure the correct use of generative AI and make it a valuable tool for all employees, ultimately enhancing business operations.

AI assistants with governance frameworks 

Mr. Forrest Chai, Chief Information Officer at Hang Seng Bank Limited, also shared his years' worth of knowledge to the forum's audience, having started as a developer in Web technologies in 2000 and developing the first few generations of internet banking services for HSBC group, across retail and wholesale.  

Forrest said he sees the significant potentials from AI co-pilots. One of the low-hanging fruits mentioned during the forum is the Data Co-pilot, which provides banks a new way to work with their data & analytics capabilities, from enabling quick insights and complex queries that take weeks of manual labour to produce, to enabling a more cost-effective migration strategy to data cloud. This approach facilitates experimentation, enable teams with real-time data insights, and promotes data literacy across the organization.

Forrest highlighted that, as a technologist, he’s tasked with the mission to ensure that banks spend their energies and focus on “high impact” use cases. Hang Seng Bank Limited underwent a rigorous selection process, narrowing hundreds of AI use cases down to a few patterns based on justifiable return on investment and potential scalability. 

He added that scaling AI solutions in banks may experience hurdles related to governance and control. Governance frameworks are key to managing these risks, including rules, model risk, and operational resilience. These frameworks cover model risk assessment, legal compliance, cybersecurity, and regulatory compliance. He highlighted the collaborative efforts between banks, technology firms and regulators, to explore the possibilities of AI adoption, while ensuring fully compliance with regulations.

Impactful, responsible AI implementation

Mr. Alfian Michael Sharifuddin, Managing Director & Head of Technology and Operations for Hong Kong and Mainland China at DBS Bank (Hong Kong) Limited, discussed his perspectives through the lens of his experience in defining and implementing large group-wide technology and operations strategies for financial organizations.

Alfian emphasized the power of large-scale crowdsourcingⁱⁱ in identifying impactful AI use cases. DBS Bank gathered around 200 AI use cases and, through debates and voting, selected the 14 most promising ones. Among the interesting relevant cases within DBS Bank, is how they deploy an AI agent to assist relationship managers (RMs) during customer calls, providing regulatory compliance insights, and providing training recommendations for RMs.

To ensure robust and responsible AI implementation, DBS Bank has established the PURE (Purposeful, Unsurprising, Respectful, and Explainable) framework, which guides the evaluation of models for fitness of purpose and considers risks such as hallucination, overconfidence, and data privacy. The bank fosters a culture of diverse model exploration through their data chapter, consisting of AI and data scientists who analyze data sets and associated risks. Different layers of data segregation and infrastructure are employed based on data availability and sensitivity, utilizing public, or private cloud infrastructure.

Risk management is a top priority for DBS Bank, and they developed the ISOLATED framework to evaluate models. This comprehensive framework addresses inadequate human oversight (I), sensitive data use (S), overconfidence (O), log access (L), accuracy (A), toxicity (T), explainability (E), and data storage (D). According to Alfian, the bank came up with this framework to make sure that risks are managed, with an assessment of each aspect done, before models are approved for deployment.

Building trust and confidence in AI models

Based on these discussions, it is apparent that the financial sector is eager to find ways to adapt and embrace generative AI, recognizing the transformative potential of its technologies, while ensuring compliance with regulatory requirements. The speakers share a belief that proactive corporate accountability must be held to ensure that AI is explainable, transparent and fair – emphasizing the vital role of careful selection and implementation of AI models to build trust and confidence, in their application within the industry.

Model selection matters as there is no one-size-fits-all approach. IBM believes in an open, multi-model approach that enables organizations to select models optimized for their specific use cases, that deliver higher performance, better value and greater sustainability as businesses scale.

To summarize, there are three steps that organizations need to take to truly see the benefits of AIⁱⁱⁱ:

1. Start from a trusted base model - Organizations need a transparent base model that allows for trust and understanding of its training data source, model weighting, and components. This solid foundation is necessary for safe and meaningful development. This a large part of the reason that IBM has open-sourced 18 of its Granite models, under an Apache 2.0 license, including its code models, time-series, language, and geospatial models, making coding as easy as possible — for as many developers as possible. All models were trained on data that was collected in adherence with IBM’s AI ethics principles and with the IBM legal team’s guidance for trustworthy enterprise use.^iv
2. Create a new representation of your data - Organizations can build a data representation based on the selected base model to address their pressing issues. The rise of foundation models has led to new AI applications in areas like payroll, social media, and internet search. To facilitate this, IBM and Red Hat introduced InstructLab, an open-source project that builds on the LAB (Large-Scale Alignment for ChatBots) technique for a community-driven approach to language model development through skills and knowledge training, enabling incremental learning for foundation models, empowering users to feed new data to the model and enhancing its abilities without starting from scratch. IBM's Granite-7B language model is integrated into InstructLab, allowing the addition of new skills and knowledge, without losing previous training.
3. Deploy, scale, and create value with your AI - Once a business' data is compatible with a model, they can deploy, scale, and generate value through their AI applications. For instance, IBM watsonx.data enables enterprises to scale AI and analytics with their own data wherever it resides. It is a core component of the IBM watsonx AI and data platform, allowing enterprises to create custom AI applications for their specific business needs, access and manage all data sources and accelerate the implementation of responsible AI workflows—all on one platform.^v

To help enterprises including those from the highly regulated industries to accelerate and scale generative AI against business and IT complexity, at our annual THINK conference, IBM announced several new updates to our watsonx platform, as well as our upcoming data and automation capabilities, designed to make AI more open, cost effective, and flexible for businesses.  IBM is delivering the next class of automation assistants and tools, powered by its Granite LLMs, that help automate key tasks such as software coding, and new IT automation capabilities to modernize IT processes, reduce IT complexity, and accelerate productivity in business workflows, with enterprises like banking with complex infrastructures, systems, and technologies to automate their work with AI-powered automation.^vi

Only by fulfilling these steps can organizations unlock the true business value and harness the transformative power of generative AI in a responsible and sustainable manner, all while scaling its impact through trustworthy data and governance.

We’d like to thank our three guest speakers for participating in our forum and sharing their valuable insights. Their expertise has fueled our discussions, and we look forward to more opportunities to continue our collaboration, exploring and advancing the potential of generative AI with trusted data and governance. Together, we can drive innovation and shape a future where generative AI positively impacts the banking industry and beyond.

Mr. Alfian Michael Sharifuddin (Left 1^st in the photo above), Managing Director & Head of Technology and Operations, Hong Kong and Mainland China, DBS Bank (Hong Kong) Limited; Mr. Forrest Chai (Left 2^nd), Chief Information Officer, Hang Seng Bank Limited; and Mr. Kenny Au (Left 4^th), Acting Head of Operations Division, the Bank of East Asia, Limited, joined the Panel Discussion of “Scaling the Impact of Generative AI with Trusted Data and Governance”, hosted by me (Left 5^th) as Data & Technology Transformation Leader of Consulting, IBM China/Hong Kong. The three guest speakers joined Mimi Poon (Left 3^rd), General Manager, IBM Hong Kong and me for a group photo after the panel discussion.

About the author:

Chris Leung, Data & Technology Transformation Leader of Consulting, IBM China/Hong Kong Limited

@LinkedIn

Chris Leung serves as the Data & Technology Transformation Service Line Leader at IBM Consulting, IBM China/Hong Kong Limited.   He is a seasoned expert in driving business transformation and technology incubation by leveraging advanced Data & Analytics (D&A) capabilities. This encompasses Generative AI, Recommender Systems, Advanced Analytics, and Big Data Platform technologies. He is widely recognized as a Trusted Advisor for Digital Product Innovation, with an exemplary track record of success for leading financial service organizations.  As a passionate proponent of responsible AI implementation and operationalization, he regularly represents IBM Consulting as a speaker at industry conferences and universities to share IBM thought leadership and his deep expertise across a wide spectrum of D&A domains.