Exploring the role of generative AI in banking - KPMG Australia (2024)

| Timing| Bias | Cost | Key genAI themes |Article series | Download | Contact

Rarely has a new technology garnered so much attention. When ChatGPT launched in late November 2022, it took just 5 days to attract over 1 million users. Consumer media ran with the story and executives spent the holiday season playing with it.

By January 2023, ChatGPT was estimated to have reached more than 100 million monthly active users.¹ Bankers poured back into the office with dreams of massive productivity improvements and – perhaps – a little more free time.

However, it's worth taking a step back from the hype to really understand what genAI is, what it can do, and the risks and opportunities involved. It is not the silver bullet to every banker’s problem. It is a key part of society in which people and machines work together.

At this point in its development, genAI can be categorised into 5 main buckets:

1. Content generators: GenAI can be used to write routine documents and communications, such as emails or announcements, or to generate images. For example, a bank may use a content generator to create the first draft of a marketing email, thereby allowing bank employees to focus on editing rather than writing.
2. Information extractors: Bankers might use genAI to summarise and analyse larger documents, in short and long-form – this could be particularly useful when looking for anomalies across a range of similar documents like employee benefits contracts, for example, or supplier terms and conditions.
3. Smart chatbots: This would be an upgrade to the traditional narrowly programmed chatbots most banks use today, using genAI to enable conversational interactions that are based on a defined logic. They will likely be deployed internally at first (for example, to allow employees to find out how many vacation days they have remaining), before being adopted into higher-risk client solutions.
4. Language translators: GenAI tools make fantastic language translators and are able – to with a high level of reliability – translate and build new content in multiple languages simultaneously. This should allow banks to add a new level of personalisation to their client and employee communications.
5. Code generators: Somewhat related, genAI can convert text and natural language into code for various programing languages. This could be particularly useful for banks seeking to update their old Common Business Oriented Language (COBOL)-based programs using new-generation coding skills or for those seeking to enable low-code/no-code business solutions.
   

Banks are innovative, but they are also (rightfully) risk averse. For now, most applications of generative AI and large language models (LLMs) that you may have seen in banks have been limited to lower-risk internal purposes.

Some chatbots have been deployed to manage employee queries about product terms and conditions, for example, or to provide details on employee benefits programs. KPMG professionals have helped banks pilot genAI as information extractors to find anomalies within contracts or flag potentially fraudulent transactions. GenAI has also been used to quickly create bits of code that allow legacy systems to interact with new technologies.

There are thousands of potential use cases for genAI in banking. A recent report, The generative AI advantage in financial services, by KPMG in the US, found that 76 percent of executives plan to use generative in fraud and prevention, 62 percent in customer service and personalisation and 68 percent in compliance and risk.² Yet, for now, most banks are moving cautiously – using internal experimentation to learn iteratively and building proof-of-concepts around external use cases.

KPMG in the US
The generative AI advantage in financial services
(August 2023)

To be clear, banks have every reason to be cautious when it comes to AI – generative AI in particular. The biggest concern is around bias. AI is only as good as the data it is trained on, and if the model receives biased data, it will make biased decisions. Large language models and generative AI systems are trained on huge amounts of data which leaves significant room for bias to creep in.

Tied to this are concerns around explainability. AI often operates within a ‘black box’ that even its developers don’t always understand, which means that the decisions being made by the model aren’t always auditable or explainable. True: you can ask genAI to justify its decision, but those justifications are also based on a generative model, creating an eternal loop of inexplicability. Until the explainability piece is solved, banks are likely to remain hesitant to unleash AI in some higher-risk areas.

There are two cost buckets to consider. The first is the implementation costs — building out new apps, training them, integrating them into existing systems, testing them, putting them into production and so on. That all takes massive amounts of computing power, loads of data and access to highly skilled people. Centres of excellence may help balance that cost in the initial phases but will likely slow adoption in the long run.

The second cost bucket is operational. As much processing power, computing and energy as it takes to create a model, it takes multiples of that to maintain it. Spin up thousands of different models across the enterprise and the costs rapidly multiply (as do carbon emissions). While the efficiencyof existing models is rising and the cost of deploying LLMs is dropping, the market continues to see newer, larger and more capable models being deployed.

Some might suggest that headcount reductions will rapidly offset these costs. Yet this might be a somewhat optimistic view. Everyone thought desktop computers would lead to mass redundancies. But they didn’t. Nor did spreadsheets, process automation or email. What they did do, however, was allow people to focus on the more value-adding parts of their jobs. GenAI is expected to do the same.

Globally, KPMG banking and technology professionals have been working with clients to discuss the opportunities, risks and implications of genAI. Three key themes continue to resonate:

1. Tackle your biggest problems. Don’t implement genAI for the fear of missing out. This is about helping the business address big problems – speed to market with new products, for example, or risk processes. Understand the business outcome you want to achieve and then consider how you can use genAI to help solve those problems.
2. Go for transformation. All hype aside, genAI is creating fundamentally new approaches and models that can have a truly transformative impact on banks. Executives should be looking for big impacts at an enterprise level rather than focusing on siloed use cases and productivity gains. This is an opportunity to fundamentally rethink the way the business operates.
3. Right now, experimentation and innovation are critical. Given the pace of change, this is no time to fall behind. But it must be done within the framework of smart governance and risk management. The desire to move quickly must be balanced against the risks. GenAI can help unlock massive benefits, but only when it is applied smartly, responsibly, and holistically.

• Can technology create value in asset management?
• How AI could transform the insurance industry
• The impact of generative AI on private equity firms
• The value of digital transformation in financial services
• Frontiers in Finance – all editions