What the World's Best AI Bank Gets Right (That Most Businesses Don't)

Every week I talk to business owners who want to "do AI." They want it quickly, ideally with immediate ROI, and they want it to feel like the companies they read about in the news. DBS Bank is one of those companies. But what their story actually looks like up close is rather different from the hype — and far more instructive.

DBS is a Singaporean bank. In 2009, it had the worst customer satisfaction scores of any bank in Singapore. Staff joked internally that the acronym stood for "Damn Bloody Slow." Credit card applications took up to 30 days. Queues at ATMs were an embarrassment. Today, it runs over 1,500 AI models across 370 use cases, is projecting more than SGD 1 billion in economic value from AI in 2025, and has been named World's Best AI Bank by Global Finance.

That's a compelling arc. But here's the part that gets left out of most headlines: it took fifteen years, and the first five had nothing to do with AI at all.

Phase 1: They Fixed the Boring Stuff First

When incoming CEO Piyush Gupta arrived in 2009, his first move wasn't to hire data scientists or invest in machine learning. It was to fix broken processes and upgrade creaking technology infrastructure — work that was unglamorous, slow, and essential.

One of the most interesting things DBS did in this phase was introduce a metric called "customer hours" — a single unit measuring how much time customers wasted waiting on the bank. They set an initial target of eliminating 10 million customer hours. They ended up eliminating 250 million. That metric gave every single employee in the organisation a shared language and a shared purpose. It also built the culture of measurement that later made AI deployment possible.

Alongside this, their CTO consolidated legacy systems across 12 markets, moved to cloud and open-source architecture, and implemented API-first design principles. None of this is exciting. All of it was necessary.

Phase 2: They Decided Their Competition Wasn't Other Banks

In 2014, Gupta and his CTO visited Google, Amazon, Netflix, Apple, LinkedIn, and Facebook. They came back with a realisation that reframed their entire strategy: their real competition wasn't other banks. It was technology companies. Banks were just one type of financial intermediary — and if they didn't start thinking like tech companies, they'd be disrupted by ones.

DBS coined the acronym GANDALF — Google, Amazon, Netflix, Apple, LinkedIn, Facebook — and declared their ambition to become the "D" in that group. A world-class technology company that happened to do banking.

This sounds like corporate vision-statement fluff. In practice, it drove real structural change. They launched over 500 design thinking initiatives focused on customer journeys. They established a "Dare to Fail" awards programme, explicitly recognising well-executed experiments that didn't pan out. They ran hackathons specifically for employees over 40 to signal that transformation was for everyone, not just the technical hires. And they reorganised from vertical functional silos into cross-functional squads focused on end-to-end customer experiences.

Phase 3: AI at Scale — And Why It Worked

By 2018, DBS had clean data infrastructure, a culture of experimentation, and cross-functional teams who knew how to move fast. At that point, AI adoption accelerated dramatically. They went from 600 deployed AI models in 2022 to 1,500+ by 2025, with economic value growing from SGD 180 million to a projected SGD 1 billion over the same period.

Two things stand out about how they got there.

First, they built AI capabilities internally rather than buying them from vendors. Most banks went the vendor route — faster to deploy, lower upfront cost. DBS took the slower, harder path of building in-house. The result is that they can iterate faster, integrate more deeply, and own their competitive advantage rather than sharing it with whoever else is using the same vendor platform.

Second — and this is the one that often gets overlooked — they deployed time-to-value dropped from 18 months to 2–3 months per model. That improvement didn't come from better AI tools. It came from the infrastructure, the culture, and the organisational structure they'd spent years building. The speed was a lagging indicator of earlier decisions.

The PURE Framework: Governance as a Competitive Advantage

One of the more counterintuitive parts of DBS's story is that their governance framework — normally the thing that slows innovation down — became an accelerant.

Early in the AI journey, they created something called the PURE framework to govern how customer data could be used. PURE stands for Purposeful (data collected for a concrete reason that improves customer lives), Unsurprising (customers shouldn't be shocked by how their data is used), Respectful (data use must respect customer preferences), and Explainable (AI decisions must be explainable to customers and regulators).

By establishing clear principles early, DBS's teams didn't have to pause and debate ethics every time a new use case came up. The guardrails were already in place. This is what let them move fast — not a lack of governance, but governance that was thoughtful, clear, and embedded from the start.

Workforce Transformation: It's Not Just About Hiring Data Scientists

DBS trained more than 18,000 employees in data management. 8,000 completed their "Data Heroes" programme — not a technical qualification, but a practical programme teaching staff how to ask data-driven questions and interpret analysis. They also made a company-wide generative AI curriculum available to all 41,000 employees.

The pattern here is deliberate: they weren't trying to turn everyone into engineers. They were trying to build a workforce where everyone could work alongside AI tools intelligently — asking better questions, interpreting outputs critically, spotting when something looked off.

For a smaller business, the equivalent isn't a training programme. It's giving your team time to actually use the tools, creating space for people to report when AI outputs are wrong, and making sure the person pressing "send" on an AI-drafted email has actually read it.

What Does Any of This Have to Do With an SME in Ireland?

On the surface, not much. DBS spends approximately USD 1 billion per year on technology. They have 41,000 employees. They operate across 19 countries. Most of my clients are running teams of 5 to 50 people with a fraction of that budget.

But the underlying logic translates completely:

• Start with foundations, not features. Before you automate anything, you need to know what your processes actually are, where your data lives, and which problems you're actually trying to solve. Messy inputs produce messy outputs — no matter how good the AI.
• Patience is a strategy. DBS spent five years building the infrastructure that made AI adoption possible. You don't have five years — but you do need to resist the urge to skip straight to the shiny tool before the groundwork is done.
• Culture matters as much as technology. The businesses I see get the most from AI aren't the ones with the biggest budgets. They're the ones where the team is curious, willing to experiment, and not punished when something doesn't work on the first attempt.
• Governance isn't the enemy. Understanding what data you're using, why, and how it will be handled isn't just a compliance requirement — it's what lets you move confidently rather than nervously.

DBS is a particularly well-documented example because of the Harvard Business School case study that was written about them, and the various academic analyses of their transformation. But the same arc — fixing foundations, building culture, then scaling — shows up in almost every successful AI adoption story I've seen, at every scale.

The ones that skip steps one and two tend to spend a lot of money and end up frustrated. The ones that do the boring work first tend to find that the exciting stuff goes much more smoothly than they expected.