The trust: How Singapore’s largest bank builds AI with confidence

Three years ago, if you wanted to understand a complex topic, you'd spend hours clicking through search results, reading articles, and piecing together your own synthesis. Today, you can have a conversation with an AI that learns what you need as you go. That shift, as simple as it sounds, has led to sweeping changes across industries almost overnight.

For banks, the implications run even deeper. AI isn't just changing how employees find information — it's transforming how they serve customers, evaluate trades, process documents, and make decisions. And unlike many industries where mistakes can be corrected with a return or refund, banking operates in an environment where getting things right the first time is mandatory.

What's driving the sense of urgency around AI in banking right now?

Nimish Panchmatia: I don't think AI is an option anymore. It's squarely in our face. People who choose to ignore this will get left behind. For companies that embrace it, they'll most likely end up with a competitive edge. There's no first prize for being first, but there is a last prize for being last.

Before November 2022, nobody could have dreamed of having the kind of conversations you're now having with a bot. The technology has solved problems like hallucination and accuracy very quickly. Now with agentic AI coming into play and putting runtimes into workflows, it gives you a whole different set of possibilities for day-to-day work.

With AI evolving this quickly, how do you decide which projects to pursue?

There's no recipe. You've got to believe the technology has benefits, then choose between things that are urgent and speculative or longer-term investments.

For speculative investments, our view is we'll give it a shot and test it. Two and a half years ago, we believed getting the right accuracy for investment advisory sheets would require a lot of engineering beyond basic LLMs. We didn't call it this at the time, but we built what we now think of as agents — we applied reasoning and chain-of-thought. Nine months later, every LLM was giving us reasoning for free. That work became redundant, because others like Google started offering it, too, but we learned a lot.

The big problem is that outside of tech companies, everybody's still thinking old school, two-year or three-year investment cycles with lengthy approval processes. But the pace of change in AI is so fast. These cycles have shrunk to days, weeks, maybe months. You can't sit on a 24-month cycle and expect returns from AI.

With that in mind, where have you turned, and where are you seeing the most success with AI? And on top of that, how do you balance opportunity against risk?

We look at returns through three lenses: revenue, cost reduction, and risk management. We see a lot of value across the bank, for example, around call center processes, operations, and augmenting salespeople, etcetera. If you can save 30% to 40% of time for a salesperson preparing for calls, they can do 30% to 40% more selling. You might not get 30% more revenue, but you certainly get a lot more than before.

We have our entire corporate bank using an AI-driven chatbot for servicing questions — not transactional or advisory, but things like "What form do I need?" or "How much does this cost?" That's customer-facing with no human in the loop, but it's limited to servicing. We're in an industry where we can't get a single thing wrong. There are serious consequences.

And yet, you’re so confident, you’ve been one of the few organizations publicly disclosing AI value. How do you measure it?

A large part is test and control. A group of people gets AI treatment, a group doesn't. That takes out the noise, and the delta is what we count as AI value.

Our CSO Assistant has helped reduce call handling time by up to 20% while demonstrating more than 95% accuracy in key tasks such as providing solution recommendations and call transcriptions.

How are you building AI capabilities across the organization?

As AI entered the next wave with generative AI, we realized talent would be extremely important. But not just any talent — we need talent that has our context and understands our processes.

You can find a PhD in data science, bring them in, and they're very smart. But if they don't understand your context, how you work, and your culture, it's very hard to apply AI. AI is doing stuff humans were doing — it needs to fit in with your culture, processes, and systems.

Many companies have AI pilot projects but struggle to reach scale and get value. The problem isn't the technology. It's people, structure, and culture.

Our response was to create our “data chapter” that brings together 700 data professionals including 250 data scientists. They stay in their units for day-to-day work, but they're part of one organization responsible for upskilling, providing opportunities to work on different use cases, and creating a sense of belonging.

How do you adapt your AI approach across different markets?

Regulation drives everything. We work within different regulatory frameworks across markets — what you can do in Singapore often can't be done in Indonesia, and vice versa.

Banking operations are similar enough, though, that one use case often scales across jurisdictions. That's how we've been able to scale. But there are unique needs. Completing KYC — know-your-customer — compliance, or applying AI and machine learning, doing things like that in India or China is different from Singapore, so you need solutions specific to each market.

What we try to do is make the back-end agnostic in terms of which platform and models to use. We provide services through APIs through platforms such as Vertex AI, so it's agnostic whether we're on one model version or another.

What advice would you give other financial institutions looking to accelerate their AI journey?

You need buy-in from the top. If your CEO doesn't believe this needs to get done, it's an uphill battle. Top management needs to understand what AI can actually do and how capabilities apply to your processes.

The last thing I'd say: the tech is the easiest bit. The processes, the structure, the people, the culture — that's the hard bit. Focus on that. Most times people think, "We’ve got this beautiful technology, let's give it to everybody and things will get solved." It doesn't work like that. You’ll need to drive an organisation wide transformation and cultural shifts in order for AI to succeed. This is what DBS did, and what we continue to do, as we transform the bank and operating models, bringing our employees along on our AI journey.