Generative AI in banking, one year on from ChatGPT

ON NOV 30, 2022, OpenAI launched ChatGPT, which captivated the world with its ability to produce humanlike text responses. The event ignited a surge of interest in generative artificial intelligence (AI), prompting organisations to explore how the technology could transform their operations. Banking executives have come under pressure to incorporate generative AI into business processes. Already, many leading firms are actively experimenting and laying the foundations to operationalise and harness its capabilities at scale.

Generative AI’s potential seems boundless, but the truly transformative applications in banking are still unknown. Looking ahead, one probable future is where banking value chains would remain familiar but banks would focus on applications, crafting a multitude of innovative generative AI use cases to streamline user experiences for both consumers and employees. In this vision, generative AI would become an integral, largely unnoticed component of daily life where users are “trained” to tolerate its limitations.

Today, most banks are still in the early stage of deploying generative AI and have yet to make significant adoption progress. Still, we have seen leading players proactively explore opportunities. Their immediate focus has been on developing lower-risk, internal-facing applications that demonstrate tangible productivity benefits in daily operations.

Early examples of this include generative AI-powered “co-pilot” tools designed to augment employee decision-making processes, particularly within the back and middle office functions. Thus far, the preference has been for off-the-shelf solutions with simple contextualisation and customisation options. These solutions are scalable and cater to various departments.

Companies already have realised significant benefits. For instance, Marsh McLennan deployed a proprietary generative AI assistant called LenAI, and early adopters reported saving an average of eight hours per week when using the tool. We anticipate mass uptake in key functions in the coming years — both ready-to-use co-pilot tools and more tailored productivity boosters for specific banking functions.

Leading banks are also looking beyond individual use cases and focusing on big wins that could transform operating models. We anticipate the emergence of increasingly sophisticated applications that harness the power of both generative AI and conventional predictive AI or machine learning technologies. This integration will unlock a new realm of capabilities and propel AI’s potential within financial services.

Already, through Oliver Wyman’s work, we have witnessed some impactful pilot applications. For instance, a global bank is leveraging a combination of AI technologies to empower its risk and compliance managers. This solution automatically keeps these managers informed about the rapidly evolving regulatory changes in real time and in a manner that aligns with the expertise and decision-making processes of compliance officers. In another example from a client, a bank has deployed a surveillance tool that combines generative AI and reinforcement learning. The tool monitors employees’ social media marketing activities, flagging suspicious or unauthorised behaviours that may result in regulatory censures.

Immediate actions for banks

Banks are at a crucial juncture for determining where and how to begin their AI journey. Understanding the urgency, here are four actionable steps banks should adopt today:

Set priorities based on learning and experiments: With the surge in interest in generative AI, banks are often inundated with potential applications. Traditional prioritisation methods, conducted without a practical grasp of the technology, can lead to misguided decisions. To identify impactful opportunities, banks must build a clear understanding of generative AI’s capabilities and limitations. They must also adopt a practical approach to rapidly test feasibility through actual use cases, enabling them to assess risk and validate business cases. This approach ensures that investments are placed strategically.

Strike the right balance between risk and governance: Inaction is not the low-risk option when it comes to generative AI. Banks must establish robust risk and control frameworks, setting guardrails grounded in a comprehensive understanding of AI risks and consequences. Existing risk management processes, focused on traditional model risk metrics, should be overhauled to manage risks emerging from generative AI. Organisations must recognise the long-term impact of generative AI on their standing and prioritise managing reputational and legal risks.

Accelerate technology readiness through a multi-speed approach: To expedite the process, banks must adopt a more balanced, multi-speed approach, facilitating safe and fast experimentation for key teams tasked with building prioritised use cases in the immediate term. At the same time, they must ensure their technology infrastructure and proprietary data are ready for the longer term to power future modelling requirements with flexibility, while also prioritising security and privacy aspects.

Upskill talent to harness productivity gains: Demand for generative AI skills is expected to outpace the available talent pool. To bridge this gap, organisations must proactively institute upskilling initiatives.

The future for banks in generative AI promises transformative potential. Organisations must then swiftly identify impactful use cases, balance risk and governance, accelerate technology readiness, and rapidly upskill talent.

The writers are from Oliver Wyman. Na Zhou is partner, data and analytics lead, Asia Pacific, and Edmund Lim is principal.