Exclusive: OCBC’s 20-year data & AI innovation journey

When it comes to data, OCBC is playing the long game.

Donald MacDonald, Head of the Group Data Office at OCBC, has spent years steering the bank's approach to advanced analytics and artificial intelligence. His team sits at the heart of the financial services group, which includes Great Eastern insurance, OCBC Securities and Bank of Singapore, as well as operations across Singapore, Malaysia, Hong Kong, China and Indonesia.

"The Group Data Office is the centre of excellence for advanced analytics and AI across OCBC Group," he said. "I own the data asset, the data platform, and the data strategy, but most of my time is spent on value creation - how we use that data to drive revenues, improve productivity or reduce risk."

Foundation

For MacDonald, the key to scaling AI has been years of investment in infrastructure. "What's holding most companies back is that they haven't invested in the data," he explained. "At OCBC, we spent the last 20 years doing that. Our data is well sorted. It's then just saying, how are we going to apply it to drive the most value?"

This long-term strategy underpins how his team decides which projects to pursue. "There are thousands of potential use cases for AI in a bank, so it's about being selective," he said. "We look at expected return, cost savings, effort required, reusability, and whether it's an ethical use of data."

Compliance and risk teams play a dual role. "They're both a customer and an oversight function," he explained. "We follow their policies when building models, but we also provide services for them, from anti-money laundering to fraud detection."

Cloudera at the core

A central part of OCBC's journey has been its decade-long partnership with Cloudera. "In 2015 we realised traditional data warehouses were too costly and slow for our ambitions," MacDonald said. "Cloudera and Hadoop let us bring in unstructured data at a fraction of the time and cost."

He estimates that integrating data through Cloudera is "about 10% of the cost" compared with older systems. "In our first ten years we brought in 50 systems. In the last ten years, we brought in more than 300," he added. Today, both OCBC's data and machine learning sit on Cloudera.

The benefits go beyond cost. "Setting up Cloudera infrastructure used to take four months. Now it can be done in one hour," he said.

On-premise over cloud

Despite the rise of cloud services, OCBC keeps most of its sensitive data and AI workloads on-premise. "We have 30 plus Gen AI applications running in production today," MacDonald explained. "A lot of those use very sensitive information, and we don't want that going outside our firewall."

Security is one reason, but cost is another. "For one contact centre use case, if we'd run it in the public cloud it would have cost us $3 million in API calls," he said. "We have 30 such use cases. On-premise GPUs and open-source large language models make it far more affordable."

Innovation also plays a part. "Cloud giants release something, but within two to three months the open-source community catches up. For us, the models are good enough."

Real-time and responsible AI

Only 20 of OCBC's 350 systems run in real time, reserved for cases where milliseconds matter. "Not every use case needs to be in real time," MacDonald explained. "But for things like scam detection, we process payment transactions in less than 15 milliseconds. If we identify a likely scam, we can stop the money before it leaves the account."

The bank's AI strategy extends to generative tools, designed for both universal use and specific roles. Staff have access to OCBC GPT for content, Buddy as a knowledge assistant, and Document AI for translation and summaries. Copilots are tailored to teams, such as Wingman for coding and Holmes for sales, which processes 200,000 pages of research daily to provide talking points.

One of the most striking examples is in private banking. "Onboarding used to take 10 days and require 40 documents per customer," he said. "Now our agentic solution does it in less than an hour, while improving quality and standardisation."

From months to days

MacDonald admits implementation speed depends on complexity and integration needs, but thanks to two decades of groundwork, progress is rapid. "If we have the data, most use cases can be built in less than a month," he said. "That's very fast compared to 20 years ago when projects took months and everything was ad hoc."

The productivity gains are tangible. "For developer productivity we quote a 20% uplift," he said. "Writing code is only part of the job, but AI tools like Wingman make a clear difference."

OCBC's approach reflects a clear philosophy: build the foundation, centralise where possible, and apply AI responsibly to drive value. For MacDonald, the lessons are simple but hard-earned. "What we can do today compared to ten years ago is not comparable," he said. "It's a different world."