Transforming the banking experience with genAI

AI has transformed multiple sectors.

Within banking, for instance, AI has reshaped customer service and operational efficiency. The recent emergence of generative AI (genAI), a subset of AI that can create content and generate human-like text, has further enhanced these changes.

The research arm of consultancy firm McKinsey estimates that genAI can add between US$200 billion and US$340 billion in value to the global banking sector every year through increased productivity. This translates to an uptick in revenue of between 2.8% and 4.7%.

“It’s like the world is waking up, understanding now what AI can do for them. And that has led to huge demand in the business,” says Donald MacDonald, head of group data office at OCBC / Photo credit: OCBC

The use of genAI will go far beyond contributing to the bottom line too. It could also enhance customer satisfaction, improve decision-making and employee experience, and decrease risks through better monitoring of fraud and risks.

This is why OCBC is currently riding the genAI wave and has been one of the early adopters of AI among Singapore banks.

The AI journey at OCBC

The bank first set up its AI Lab unit in 2017 and extensively uses AI across three areas: generating revenue, improving productivity, and reducing banking risks. Today, the AI unit has expanded exponentially and is known as the group data office.

AI has also become embedded in the bank’s day-to-day operations – from personalizing credit card deals to detecting fraud.

OCBC was the first Singapore bank to roll out a genAI chatbot, OCBC GPT, in November last year. The chatbot assists the bank’s 30,000 employees globally with writing, research, and ideation.

Donald MacDonald, head of group data office at OCBC, says that while traditional AI is focused on rational decision-making, genAI focuses on creativity.

In this regard, the technology is “transformational,” he says. In the past, every problem statement would require OCBC to build separate AI models using traditional AI. For example, the bank would have to build three models to generate, document, and conduct unit tests for computer coders when helping businesses solve problems.

GenAI, however, requires only one language model to execute all three tasks.

“The models are also very good at what we call zero-shot learning, which is the ability to solve a task without being trained specifically how to do it. These models allow us to build things much, much faster than we could in the past,” says MacDonald.

But the biggest benefit of genAI, he says, is that it makes AI more accessible.

“It’s like the world is waking up, understanding now what AI can do for them. And that has led to huge demand in the business,” he adds.

OCBC’s two-pronged approach to genAI

MacDonald credits OCBC’s quick adoption of genAI to its early implementation of AI technology.

“One of the reasons I think OCBC has moved so fast on genAI is that we were actually using the first generation of large language models back in 2019 and 2020,” he says.

These language models include those offered by companies like Google, such as Bert and FinBert, which OCBC used to read the news and detect early signs of credit default.

When OpenAI’s large language model GPT-3 and ChatGPT took off at the end of 2022, OCBC’s team was able to move quickly to make its existing applications better, says MacDonald.

He also attributes the bank’s success in genAI to the strength of its engineering team.

All of OCBC’s AI tools are built by its in-house data platform team using open-source tools.

“When you build (the model) yourself, you know exactly how the algorithm is coming up with those predictions. This gives greater transparency on how decisions are being made which you can better explain to the regulator or customers,” he explains.

The GenAI tools deployed within OCBC are used by its staff around 500,000 times each month.

MacDonald says that the bank has a two-pronged approach to using genAI within the bank.

The first prong involves designing universal tools to help every staff member work more efficiently. OCBC GPT, for instance, can help employees draft content, such as their performance appraisals.

There is also Buddy, an internal knowledge assistant that can answer employees’ questions on internal policies such as parental leave or expense application.

The second prong involves integrating specific AI tools with key functions within OCBC to help each role within the bank operate more efficiently.

For example, the IT team at OCBC has Wingman, an AI tool that helps them code faster. Meanwhile, the private banking relationship managers have a new genAI tool that helps them quickly get insights and talking points on investment ideas.

MacDonald says that the use of GenAI within OCBC is currently focused on improving the productivity of staff. There is no direct interaction between customers and genAI functions as of yet, as there are still some risks involved.

Hallucination, which refers to AI providing customers with incorrect information, is one of these issues. There is also the risk of adversarial prompting, where AI could reveal workings of the bank’s models that OCBC does not want external parties to see.

Focusing the use of genAI internally helps the bank build up confidence and capabilities in AI, says MacDonald.

Once the bank has developed guardrail controls that are robust enough to minimize risks, it will consider allowing genAI to be used directly with customers.

Ultimately, OCBC’s customers still benefit, even if genAI is currently limited to internal use.

“The employee has a lot more information at their fingertips and is much more updated because genAI is able to pull information together. So I think our customer-facing employees are able to serve their clients better with the use of these AI tools,” says MacDonald.

Envisioning a new future for banking

In the past year alone, genAI has made “huge strides,” says MacDonald. Hallucination risks have lessened, while OCBC has also improved its own capabilities to detect incorrect information generated by AI.

GenAI has also moved beyond chatbot-type functions to handling complex tasks. For example, it can now read and understand images and documents and automatically draft a proposed response.

“So that really extends the capability of generative AI to actually be much, much more productive,” says MacDonald.

GenAI will create a future where customers can use conversational language to interact with banks and receive hyper-personalized service.

He imagines a day when a customer can verbally ask OCBC’s AI tools for updates on the market, and the AI tools can recommend and execute investments based on the customer’s risk profile.

Given the progress that OCBC has made in genAI over the past year, MacDonald is confident that his imagined scenario can materialize within the next few years.

“All the building blocks required to achieve that vision are already there today. It’s just about knitting them together and making sure you have the guardrails in place to allow you to be able to execute on that.”

This story was republished with permission from The Business Times. It was moderately edited to reflect Tech in Asia’s editorial guidelines.