While there is currently a lot of buzz around generative AI like ChatGPT, using the technology will become the norm in the near future. Many banks are already testing or have deployed generative AI at scale, with the common goal of enhancing employee productivity and boosting operational efficiency.
United Overseas Bank (UOB) is trialling Microsoft 365 Copilot with 300 employees for a year. Microsoft 365 Copilot combines the power of large language models with business data from users’ calendars, emails, chats, documents, and meetings, as well as Microsoft 365 apps like Outlook, Word, PowerPoint and Teams. As such, it can serve as a personal assistant since users can ask questions or instruct it to execute a task in natural language. For instance, UOB employees can get Microsoft 365 Copilot to summarise email threads on Outlook, retrieve and reference information within the bank, or transform raw data into visualisations in Excel.
Meanwhile, OCBC has developed and deployed its generative AI-powered solutions across the organisation. This includes OCBC Whisper, a speech-to-text technology that automatically examines every sales conversation with customers to spot anomalies in the sales process. To further extend its use, the solution is currently being trialled at the bank’s contact centre to transcribe calls (with at least 90% accuracy) and summarise them to reduce the time a call agent spends handling customer calls.
As for OCBC Wingman, it helps developers generate, debug and improve computer codes automatically. This standardised code quality ensures the code does not leave the bank’s environment and saves 20% of developer effort during code-building. OCBC Wingman currently writes about half a million lines of code per day.
Given its benefits, generative AI is seen as a growth multiplier. Accenture believes generative AI could magnify a bank’s operating income by reducing costs and driving revenue growth.
“For the cost side, we predict a 9% to 12% reduction in mid- and back-office costs achieved by a productivity increase of 7% to 10% in corporate functions. On the revenue side, we anticipate that generative AI could create a 17% increase in time allocated to client interactions and advice, responsible for around 80% of banking revenue. This additional time could translate into a 9% surge in revenue,” Accenture said in a September blog post.
See also: SFF 2023 spurs discussion on generative AI and the future of finance
Bolstering defence against financial crime
Banks are also looking to implement other artificial intelligence (AI) forms, such as machine learning and deep learning. Research firm International Data Corp expects banks worldwide to spend an additional US$31 billion ($42.5 billion) on embedding artificial intelligence (AI) into existing systems by 2025.
Fraud management is a top priority for such projects. This is unsurprising as financial crime and the cost of financial crime compliance are rising. Financial firms in Asia Pacific were estimated to have spent a total of US$45 billion on compliance costs in the past 12 months, according to data and analytics provider LexisNexis Risk Solutions.
Traditionally, banks use rule- and scenario-based tools or basic statistical approaches for transaction monitoring. These tools have limited effectiveness, often failing to capture the latest trends in fraudulent or money laundering behaviour.
AI or machine learning offers a smarter and faster way of tackling financial crime as it leverages more granular and behaviour-based data, can adapt to new trends, and continually improves over time. By using AI to automate processes and conduct multi-layered deep learning analysis, banks can better identify potential fraudulent and money laundering activities with significantly fewer false alerts, considerably increasing transaction monitoring efficiency. According to McKinsey, a financial institution improved suspicious activity identification by up to 40% and efficiency by up to 30% after replacing its rule- and scenario-based tool with machine learning for transaction monitoring.
Furthermore, AI can alert banks of suspicious patterns or relationships before human experts spot them. Artificial neural networks, for example, can help banks predict the next moves of even unknown bad actors looking to exploit the loopholes in the traditional, binary rules-based transaction monitoring systems. They could also investigate linkages between customers and employees to alert banks to suspicious dealings. This is because artificial neural networks link millions of data points from seemingly unrelated databases — from social media posts to internet protocol addresses, real estate holdings, and more — to identify patterns.
Building blocks of AI
AI will play an increasingly central role in creating value for banks. But for that to happen, banks must transition from a legacy architecture and operating model to an automation and cloud-first strategy.
According to McKinsey, building the core technology and data capabilities upon a highly automated, hybrid cloud infrastructure can enable an AI-driven bank to scale rapidly and efficiently as it gains competitive and differentiating capabilities. It therefore advises banks to focus their transformation on the following areas:
Modern API and streaming architecture
Banks should integrate internal and external systems to support seamless customer journeys. This calls for a robust, scalable, and standardised approach to building and hosting integrations and application programming interfaces (APIs). The APIs should also be rigorously tested for performance and developed using agile release principles. Those efforts will allow product innovations to move from concept to production and deploy minimum viable products within 30 to 60 days.
Additionally, banks should consider building a high-speed data streaming channel to complement their robust API strategy. This will enable standardised asynchronous data transfer across the enterprise in real-time.
Core processors and systems
Banks must shift from traditional, complex, and tightly intertwined core systems to lightweight and highly configurable core product processors and workflows to be agile. Those processors are also complemented by microservices or discrete applications (such as for payment card accounts) that externalise the logic within traditional core platforms.
By moving to lightweight core processors and systems hosted on scalable, modular, and lean platforms exposed as APIs, banks can better support real-time reconciliation and make changes in live systems with zero downtime. Using modern cloud-based infrastructure to host such platforms also makes it easier to scale up. If successfully implemented, a lightweight processor platform can reduce the time to market for new products.
Data management for AI
A modern data and analytics platform is crucial to fuel the real-time machine learning models used for decision-making. The data platform must be capable of ingesting, analysing, and deploying vast amounts of data in near real-time. It must also provide lab and factory teams with scalable workbenches with AI and data science capabilities.
By doing so, teams can access relevant data sets as they develop models and deploy insights in product iterations. The infrastructure should also support the development of machine models through automated and repeatable processes.
Besides that, banks should ensure proper governance and access control. Employees can leverage self-serve, real-time data and analytics infrastructure to guide value-based planning and support daily decision-making by creating machine learning models and scorecards through a well-defined lab-factory model.
Intelligent infrastructure
To modernise the IT infrastructure, banks should consider adopting the public cloud to complement the traditional infrastructure for workloads that require resiliency, scale, and hosted or managed offerings (such as hosted databases). This is because the public cloud enables velocity through higher levels of automation, templates, and reduction of operational risks.
Cybersecurity and control tower
To defend themselves from cyber threats, banks should have a centralised control tower to monitor data, systems, and networks across their IT infrastructure, ensure boundary security and identify and rectify threats and intrusions. Before deploying assets on live systems, they must also establish a well-defined set of compliance measures for security testing and vulnerability scanning.
People matter
Deriving maximum value from AI also requires the right skills, which can be challenging with the current global AI and data analytics talent shortage. In response, some banks are upskilling their existing employees.
DBS, for instance, offers technical training such as coding and digital training programmes that teach employees how to think digitally or in a more data-driven way. The latter focuses on AI’s softer skills, such as how to use data responsibly and ask AI the right questions to get more accurate and relevant answers.
The bank also leverages gamification to encourage non-technical employees to gain technical skills, such as via the DBS x AWS DeepRacer League. The programme taught DBS employees the basics of AI and machine learning through online tutorials before challenging them to use that knowledge to program their autonomous model race car. These machine-learning models were then uploaded onto a virtual racing environment where employees experimented and iteratively finetuned their models as they engaged each other in friendly competition.
McKinsey estimates that AI technologies could potentially deliver up to US$1 trillion ($1.37 trillion) of additional value annually for the global banking industry. For that to happen, banks must look at where AI can help improve operational efficiency, reduce costs and risks, and capture new growth opportunities. They should then redesign their IT backbone and processes while ensuring they have the right skills to support their AI strategy. Combining those factors will be crucial in realising a truly agile, innovative, and resilient bank.
