DBS, the largest bank in Singapore, wanted to run like a tech start-up: fast, nimble, customer driven. Piyush Gupta, the CEO of DBS, sought to transform the organization by getting employees to use data analytics and AI to enhance the bank’s value to customers and sharpen its competitive edge.
Cross-functional teams comprising data scientists, AI experts, technologists, and business leads worked together. They developed new intuitive products and services, uncovered new ways to maintain an ongoing dialog with customers, and reinvented ways of working. McKinsey worked closely with DBS at key junctures on this journey to help bring this vision to fruition.
In this interview with Elaine Ee, McKinsey’s head of publishing in Asia, Piyush reflects on DBS’s transformation into an AI-fueled bank and its partnership with McKinsey.
DBS: Transforming a banking leader into a technology leader
Q. What advice can you give to financial institutions who are undertaking similar initiatives?
In any AI transformation, data is the heart of the whole operation. The second leg of the stool, so to speak, is developing the AI models; the third is the organizational changes that have to be made for the transformation to become the day-to-day reality.
But it starts with the data. You have to be thoughtful about how you create and architect your data lakes—or data repository—and how you provision them because there is this inherent contradiction between data privacy and access. And then you have to define the appropriateness of data use. We came up with a principle called PURE: data must be purposeful, unsurprising, respectful, and explainable. McKinsey guided us through these phases.
Finally, there’s putting in the hard yards of actually migrating the data. It’s taken us more than ten years and about a 100-person data factory just to clean up and migrate the data over time.
Q. The Harvard Business School case study on DBS highlights how much experimentation was involved and how much so-called “failure” is tolerated—but this must be challenging to do as a financial institution.
Yes, that’s true. The experimentation wasn’t in terms of the data, which is more about consistency and constancy. It’s about how you develop demonstrable use cases that show value along the way. And that’s where we spent ten years and had a number of failed attempts.
Developing and refining AI models is where experimentation came in: figuring out which ones worked, which didn’t, and where we could do more and do less. And this experimentation has to be done in pilot mode because you can’t run it full scale.
When we got to this stage, McKinsey helped us think through how to institutionalize and scale up these capabilities: how to develop models, create a library, reuse models, and establish protocols around them.
Today we have about 700 data folk, 200 data scientists, and other translators and engineers; it’s an organization we built over time. We have a library of 1,500 models. It used to take us 18 months to develop and scale a model; now we can do it in two or three months.
This deep experience in AI and structured data prepared us for when gen AI seemingly “exploded” on the scene; we were ready to start experimenting and integrating it to further advance our capabilities.
Q. Were there breakthrough moments or was this more of an ongoing evolution?
I think it was a constant evolution. Three or four years ago, we could see that wherever we were using AI, the results were beginning to come in. But it was anecdotal. We were using the AI model for hiring people, and we began to see that these people had lower attrition rates and better performance. Our AI models in marketing were producing better outcomes. This was an “aha moment” because once we started calculating the economic value across all of the models, we could see hundreds of millions of dollars, and this galvanized the whole company because they could see the value.
Q. How did you change the culture to become a data-driven institution?
First of all, we trained 8,000 people in a program called Data Heroes on how data can be used and the kinds of questions and answers that data can give.
Then we ran another training program during COVID called Deep Racer, the AWS program, which taught people how to program autonomous driving vehicles so you could race them around tracks and take part in competitions—internally, externally, then globally.
We also mandated that all internal presentations and reviews, as much as possible, be run with control towers and live dashboards. Many of our meeting rooms and operating centers now have big screens with live data. All throughout the day, people are looking at what the data is telling them. It was a major cultural shift, and it’s really powerful.
Superagency in the workplace: Empowering people to unlock AI’s full potential
Q. In the transformation, is the technology leading, or are the business and customer leading? Or is it not so straightforward?
It’s actually that everybody leads. McKinsey helped us build a new operating model around 33 platforms based on our business segments and products. Each one had a “two in a box” leadership model, which meant it was led jointly by someone from the business and someone from IT. They share KPI outcomes. We apply the same principle to all our technology and innovation work.
Today we are running the bank through 60 horizontal journeys focused on the customer experience, addressing major pain points like account opening and ATM waiting times. The leaders could be from operations, risk management, technology, any part of the bank; the journeys are made up of the people who bring the most value, and a large part of the AI and data use is done collectively by them. So, anyone could lead.
Q. With the data and tech powering the organization, how much human oversight is involved?
For the time being, we’re running AI copilots, with a human in the loop, to ensure accuracy. We also want to make sure that the client base is comfortable dealing directly with AI, so we will start experimenting with it on a consent basis. But copilots also give our people confidence in that they can improve their knowledge; their mundane tasks are reduced so they can focus on the higher-value things. So that’s been quite helpful.
In the customer relationship space, we have two programs. Our Next Best Nudge program is now processing 45 million hyper-personalized interactions with customers a month, such as “This looks like a duplicate payment; please verify it.” The Next Best Conversation program sends a nudge to the relationship manager to follow up, such as: “You might want to call Elaine about duplicate 12345.”
Q. Going forward, where do you see AI taking DBS in the next several years?
It is going to be very, very powerful. It changes productivity and the way work is done. I think ten years from now, you could get a completely AI-powered bank—a platform where each individual customer actually creates their own AI version of the bank. So, the sky is the limit in terms of where you could go with this.