How DBS became an AI bank.

The short version.

In 2025 DBS reported about S$1 billion in economic value from data analytics and AI/ML, from more than 2,000 models across 430 use cases.²⁵ DBS counts three things in that figure: extra revenue, cost savings and productivity, and losses avoided from fraud, scams and credit. It measures the value with test and control groups, and the number sits in the narrative part of the annual report rather than the audited accounts.³³

The figure is the last step of a sequence that began in 2009. DBS first took its technology back in-house and rebuilt its data centres. From 2014 it set out to be "digital to the core". In 2018 it reorganised into 33 platforms, each run jointly by a business lead and a technology lead on shared KPIs. It then built a governed data platform and an AI platform on top. The first AI value figure was published in 2022, thirteen years after the start. Between 2021 and 2023 a run of outages brought a capital penalty and a six-month restriction from the Monetary Authority of Singapore. That episode belongs in the story as much as the awards do.

Lecture slides (PDF)

Where DBS started.

DBS began as a policy bank. The Development Bank of Singapore was founded in 1968, three years after independence, "for the express purpose of financing the nation's development and industrialisation". Its first chairman, Hon Sui Sen, and part of its first team came from the Economic Development Board, and the government put in SGD 49 million of the SGD 100 million starting capital.⁷⁷ Its mass retail base came thirty years later. On 24 July 1998 the Finance Minister announced that DBS would buy POSBank, the national savings bank, in an SGD 1.6 billion deal, part of a government push for local banks to consolidate and compete internationally.⁷⁸

Before 2009 DBS had the lowest customer satisfaction scores of any bank in Singapore. Paul Cobban, later its Chief Transformation Officer, tells of a taxi driver who said the letters stood for "damn bloody slow".² Technology was largely someone else's. David Gledhill, who joined from J.P. Morgan in 2008, described it this way: "When I arrived, 85% of our work was outsourced. All of our infrastructure and applications were with IBM, and Accenture was driving our core banking programme."³ Piyush Gupta, previously Citi's CEO for South East Asia-Pacific, was appointed CEO in September 2009 and took office that November.¹ The strategy he set was to build an Asia-centric bank delivering double-digit ROE.⁴

Fixing the basics.

DBS's own label for the first five years was "Fix the basics". The foundations listed in the 2017 investor-day deck are unglamorous: build resiliency, revamp the data centres, start insourcing and build engineering bench strength, open a security operations centre and a monitoring centre.⁵ Over half the technology team changed in Gledhill's first six months.³ On the service side, the bank spent six months defining "Asian service" as RED: Respectful, Easy to deal with, Dependable. It then ran process-improvement events that Cobban says removed 250 million customer hours of waste a year.²

The spending mix shows the shift. Of DBS's build spend, 12 per cent went on digital in 2010, 29 per cent in 2014 and 56 per cent in 2017; the rest went on the core.⁵ MIT CISR researchers who followed the bank describe this phase as having "radically 'rewired' the entire enterprise for digital innovation".²⁶

The digital vision.

Gupta has said the board concluded in 2013 that "the future for us and for our industry would have to be digital. We felt that if we didn't lead the charge, frankly, we might die." The board then gave an extra $200 million, "in their words, 'to go blow it up.'"⁷ The 2014 annual report records it as SGD 200 million over three years, on top of SGD 1 billion already invested in strategic technology.⁹

The vision came under the banner "Making Banking Joyful". It had three pillars: become digital to the core, embed the bank in the customer journey, and create a 22,000-person start-up. The stated aim was to make DBS "invisible".⁴ The technology benchmark was GANDALF. Gledhill explained that the first letters of Google, Amazon, Netflix, Apple, LinkedIn and Facebook spell GANALF, "missing a D… our mission became how to become the D in GANDALF". The 2018 targets were to become cloud-native, increase release cadence tenfold, and build for APIs.⁸⁵

The culture work was built into the performance system. In 2015 running an experiment was in everyone's KPI, and the bank ran about 1,000.⁷ Hackathons were written into talent development and, in Cobban's words, "replaced the executive training budget".¹⁰² About 250 senior managers were each required to sponsor a customer or employee journey.⁴⁵¹ Hiring changed too. In February 2017 DBS announced Hack2Hire, which it called the first hackathon-based recruitment in a Southeast Asian bank: an online assessment, then a live two-day hackathon, with full-time offers for the shortlisted.⁷⁹ Even meetings got rules. Under MOJO (Meeting Owner, Joyful Observer) every meeting has an owner who runs it and an observer who keeps time and gives honest feedback, and it must start and end on time with a fixed agenda. INSEAD researchers report it saved more than 500,000 employee hours.⁸⁰ The results showed outside Singapore first. digibank launched in India in April 2016 as a mobile-only bank and signed up over 800,000 customers in nine months.¹¹⁷ That July Euromoney named DBS the World's Best Digital Bank.¹² In November 2017 DBS opened a developer platform with 155 APIs, then described as the world's largest for a bank.¹⁴

Proof that digital customers pay.

The transformation became a shareholder argument in November 2017. At its investor day DBS split its Singapore and Hong Kong consumer and SME customers into "digital" and "traditional" groups and published the economics of each.⁶ A customer counted as digital if they bought a product through a digital channel or did more than half their transactions digitally. This is the table that made "digital" a board metric at DBS.

Digital customers were 39 per cent of the base and produced 60 per cent of income and 68 per cent of profit before allowances.⁶ One caution belongs next to that. The group cost-income ratio did not fall in step: it stayed between 43 and 45 per cent from 2012 to 2019, on the bank's annual reports.⁹¹⁹ The gains were real at segment level. At group level they were absorbed by acquisitions and new investment.

Reorganising around platforms.

In 2018 DBS set out "to think and behave like big technology companies by instituting the platform operating model", in Gledhill's words in that year's annual report. "Platforms bring together people, funding, technology assets and apps to deliver a customer service or product. They are based on a two-in-a-box framework which means that platforms are co-developed and maintained by the business and its technology partners who work together on joint goals, business strategy and an execution roadmap." Thirty-three platforms were established that year.¹⁸

Gupta later credited McKinsey with helping build the 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."²⁸ Jimmy Ng, Gledhill's successor as CIO, put the consequence plainly in the 2019 annual report: "Bankers became product owners as a result, with the platform as the engine for transformation."¹⁹

How a platform is run.

The published record describes eight mechanics. None of them is technology on its own. They settle who owns an outcome, who pays for it, and how it is measured.

Two points in the table do most of the work. Funding the platform removes the annual business case for each change. Two-in-a-box leadership on a shared scorecard removes the usual argument about whether a missed target was a business or a technology failure. The journey layer added from 2021 addresses the weakness of any platform model: a customer's problem usually crosses several platforms. The IMD case records more than 60 journeys covering over 60 per cent of revenue by 2024.²⁹

The data and AI layer.

AI at DBS did not start well. In 2013 Gupta signed the bank up for an AI lab with Singapore's A*STAR; it worked on half a dozen projects, "none of which were successful".³⁰ What changed was the foundation underneath. The 2018 annual report set out the PURE test for any use of customer data: Purposeful, Unsurprising, Respectful, Explainable.¹⁸ By 2019 the ADA data platform ("Advancing DBS with AI") had onboarded close to 10 per cent of the organisation.¹⁹ In 2021 the bank fully deployed ALAN, "a single AI/ML platform that enables data and models reusability whilst ensuring 100% of use cases and models are compliant with our PURE and AI governance frameworks".²¹ A cross-functional Responsible Data Use Committee reports to the Risk Executive Committee.²⁴

The current CIO, Eugene Huang, says the platform cut the time to value for an AI or machine learning use case from 18 months to about two to three months.⁴⁹ The customer-facing output is volume. In 2022 DBS sent 45 million personalised nudges a month to about five million customers.²² In 2024 it sent more than 1.2 billion to over 13 million customers.²⁴ Generative AI arrived for staff in 2023 as DBS-GPT, and by 2025 it was available across the bank.²³²⁵ In 2026 the focus has moved to agents. CEO Tan Su Shan describes an agent registry with "accountability, observability, traceability and evaluations of the agents".⁴⁷

What the S$1 billion means.

Three things matter when a board uses this number as a benchmark. The first is what it is. It is "economic value": revenue uplift, cost avoidance and productivity, and losses averted. It is not a savings figure, and at 2026 exchange rates it is about US$770 million. The second is how it is measured. Nimish Panchmatia, DBS's Chief Data and Transformation Officer, says: "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."³³ That method is sound, but the result is not audited and no per-use-case breakdown is published. The third is that the target moved. In 2022 the S$1 billion was a revenue aspiration five years out.²² By 2024 it was an economic-value target for 2025, and the 2022 figure had been restated from S$150m of revenue to S$180m of economic value.³²²⁴ None of that makes the figure wrong. It means another organisation's number is only comparable if it is defined and measured the same way.

External rankings agree on leadership but not on everything else. The Evident AI Index placed DBS tenth of 50 large banks in 2023, sixteenth in 2024 and eighteenth in 2025. On the leadership pillar it was first in 2023 and 2024 and second in 2025. In 2024 Evident noted DBS was one of only two banks, with JPMorgan Chase, to publish a total realised value from AI, and was "dragged down by Innovation".⁴³⁵⁵ By the 2025 index, four banks reported realised returns across their use cases.⁵⁷ In October 2025 Global Finance named DBS the World's Best AI Bank.⁴⁴

How DBS compares with other banks.

The table sets DBS beside five global banks that rank high on AI and three regional competitors, using what each has published. The value column states what kind of figure each bank gives. Only DBS publishes a bank-wide figure with a stated measurement method. JPMorgan's comes from its chief executive in an interview, with no method disclosed. The others publish a single use case, a technology saving, targets, or nothing. The figures in that column are different measures and should not be ranked against each other.

Three things stand out. DBS's measurement discipline is unusual: none of the other eight describes a test-and-control method for a bank-wide figure. Its overall Evident rank has slipped as JPMorgan, Commonwealth Bank, HSBC and Bank of America invested heavily, even while its leadership score stayed near the top. And of the nine, only DBS and OCBC carry a Singapore regulator's capital add-on tied to technology or digital controls, which is the local supervisor's signal of how it prices those failures.³⁷⁷³

The outages and the regulator's response.

The same speed brought a governance failure. In November 2021 DBS's digital services were down for two days. In February 2022 MAS applied a 1.5 times multiplier to the bank's risk-weighted assets for operational risk, about S$930 million in extra regulatory capital. MAS noted "deficiencies in DBS Bank's incident management and recovery procedures".³⁴ After another outage on 29 March 2023, MAS raised the multiplier to 1.8 times, about S$1.6 billion in total.³⁵ The October 2023 failure began with the cooling system at a data centre. DBS could not fail over because of a network misconfiguration. Up to 810,000 attempts to access digital banking failed, and about 2.5 million payment and ATM transactions could not be completed. MAS counted five disruptions in eight months and told Parliament: "This is unacceptable."³⁶

On 1 November 2023 MAS imposed a six-month pause on DBS's non-essential IT changes. During the pause DBS could not acquire new business ventures or shrink its branch and ATM networks. MAS found shortcomings in "system resilience; incident management; change management; technology risk governance and oversight".³⁷ The same day the board apologised. It set up a board committee for technology risk, moved technology risk management under the Chief Risk Officer, split technology and operations into two units, and set aside SGD 80 million for resilience.³⁸ The 2023 annual report named one cause directly: "The shift from a monolithic mainframe to a cloud-native, microservices-based approach… created a more complex infrastructure requiring additional operational rigour and oversight."²³ Senior management's variable pay for 2023 was cut by 21 per cent and the CEO's by 30 per cent.³⁹ MAS let the pause end on 30 April 2024, and said the 1.8 times multiplier would stay until DBS showed it could keep services available.⁴⁰ I found no MAS notice lifting it.

The workforce.

The insourcing created the talent base. The 2019 annual report records a 24 per cent rise in technology professionals that year, to close to 6,000.¹⁹ By 2021 DBS had trained over 18,000 employees, according to MIT Sloan Management Review.³⁰ Some of that training was a game. In 2020 DBS put more than 3,000 staff through a DBS x AWS DeepRacer league, in which each learner programs a model race car with machine learning. A DBS technology director won the global AWS DeepRacer F1 ProAm event that May.⁸¹ Gupta raced too and, by MIT Sloan Management Review's account, was happy to finish in the top 100 among his staff.³⁰ The other side came in February 2025. Gupta said DBS expected to cut about 4,000 contract and temporary roles over three years as AI took over the work, adding: "In my 15 years of being a CEO, for the first time, I'm struggling to create jobs."⁴¹ Permanent staff were not affected. In August 2026 DBS committed to give all 40,000 employees foundational AI skills and identified over 11,000 for deeper reskilling in roles that AI changes significantly.⁴²

The bank it became.

By 2025 DBS was a different size of institution. In November 2020 India's government and central bank merged the troubled 94-year-old Lakshmi Vilas Bank into DBS Bank India under a statutory scheme. DBS put in INR 2,500 crore (SGD 463 million), and LVB's depositors and employees moved across on their existing terms.⁸² In August 2023 the bank committed up to SGD 1 billion and 1.5 million staff volunteer hours over ten years to low-income and vulnerable communities, at SGD 100 million a year from 2024.⁸³ In June 2025 DBS became the first Singapore-listed company worth more than US$100 billion, with a Morningstar analyst crediting much of the rise to growth in wealth management.⁸⁵

How the bank is led.

The handover was planned for years. Tan Su Shan joined DBS in 2010, built its wealth business, then ran consumer banking and later institutional banking. She was named Deputy CEO in August 2024 and succeeded Gupta as CEO on 28 March 2025, the first woman to lead the bank. DBS describes the appointment as "the culmination of a decade-long succession process", in which "a strong field of internal candidates was put through an extended development programme".⁸⁴

Both CEOs led the AI work by doing it. Gupta raced in the DeepRacer league.³⁰ Under Tan, staff had built about 26,000 personal AI agents by the end of 2025, according to Forbes Asia's reporting.⁸⁶ Her rule for running a bank through extremes: "When things are really good, don't lose your head; when things are really bad, don't lose your heart."⁸⁷

The lineage from 2009 to 2025.

Each capability below was built for its own reason and later became the base for the next. Read down the right-hand column and the AI result stops looking like an AI project.

What boards and agencies can take from it.

1. Benchmark against the sequence. DBS published its first AI value figure in 2022, after thirteen years of engineering, data and organisational work. An organisation that starts with AI use cases and skips the ownership and data work should not expect the DBS curve.
2. Put a business owner on every model. Two-in-a-box leadership on a shared scorecard is what lets DBS book AI value to a platform. Without a named owner accountable for the business result, AI value stays an estimate in the technology budget.
3. Decide the measure before the claim. Test and control is the right method. Write the definition down before the first number is published, and state it when the number is used. DBS changed its measure once, from revenue to economic value. The AI value ledger applies this grading to your own use cases.
4. Price resilience into speed. The move off the mainframe that enabled the platform model also created the complexity behind the 2021 to 2023 outages. The regulator's answer was about S$1.6 billion of extra capital and a six-month freeze. Board oversight of technology risk came after the failures, and it should come before. The technology risk board pack turns the four MAS findings into twenty questions for your own board.
5. Plan the workforce change in the open. DBS named the 4,000 roles and the 11,000 people to reskill. For a public agency the equivalent is saying early which roles change and what the retraining path is.

To see where your own organisation sits in this sequence, run the nine-question DBS sequence diagnostic.

Full timeline.

Common questions about DBS and AI.