Why most banks are struggling with AI, and what the smart ones are doing differently

Instead of running in parallel to the business, it is part of how it operates.

Banks in Southeast Asia are spending more on artificial intelligence (AI) than they have at any point in history; the ambition is real, and so are the numbers behind it. And yet, research shows that roughly half of all AI projects in banking and financial services never make it to production.

They get launched with fanfare, run as pilots, and then get quietly shelved.

The question is obvious: If the investment is going in, why are so few banks seeing returns come out? The answer has less to do with technology than most people expect.

The Southeast Asia opportunity is real, but not more than the obstacles
Southeast Asia sits at an unusual convergence of market realities. Across the region, banks are deploying AI for everything from fraud detection and credit scoring to customer service automation and real-time risk monitoring. Consumers are overwhelmingly mobile-first, generating the kind of behavioural data that makes these applications genuinely powerful.

Regulators in markets like Singapore and Thailand have generally been supportive of financial innovation. What’s more, the region has seen over US$30b ($38b) committed to AI-ready data centre infrastructure, signalling serious long-term intent from both governments and private capital.

On paper, this should create the ideal launchpad, but in practice, many banks have struggled to turn those structural advantages into commercial results. Two factors come up again and again when you look at where momentum comes to a halt: A shortage of AI talent and a patchwork of regulatory requirements that vary significantly from one market to the next.

But here’s what does not get enough attention: In many institutions, AI is still being treated as a technology initiative rather than a business one. It sits with the IT team or a dedicated innovation unit and never reaches the frontline staff and customers who would actually benefit. That is where the real problem lives.

What real progress looks like
A leading Singaporean bank generated US$565m ($726m) from more than 350 AI use cases in 2024. That figure is worth sitting with for a moment, because it reflects something specific about how they approached AI adoption. The bank did not build an impressive demo and wait for results. It embedded AI into how daily work actually gets done in credit decisioning, in customer engagement, across the business.

The banks that see the strongest returns share this characteristic. Instead of running in parallel to the business, it is part of how the business operates.

Research from Krungsri backs this up. AI-driven personalisation has been linked to revenue uplifts of around 6%. That kind of impact does not come from a standalone AI product; it is the result of systems that know enough about a customer to serve them the right offer at the right moment, built into the everyday infrastructure of the bank. The common thread amongst high performers is that AI was positioned as a seamless way of working.

The legacy trap and how to think about it differently
Digital-native banks like GXS and Trust Bank have a genuine advantage here. They were built without the weight of decades-old core systems, so they can move faster and take on more risk in how they deploy new capabilities. Traditional banks face a more demanding set of constraints. Downtime is not a real option when you are serving millions of customers across critical financial infrastructure. You cannot pause operations to modernise, and you certainly cannot afford to get it wrong.

This creates a trap that many institutions fall into. Faced with the complexity of legacy systems, they design modernisation programmes that treat those systems as problems to be replaced wholesale. These programmes tend to be expensive and long-running, creating exactly the kind of scope creep that derails them before they deliver value.

A more durable approach is to extend what already exists rather than displace it. Adding modern capabilities on top of proven foundations reduces execution risk and allows institutions to bring new products to market without betting everything on a single transformation effort. The goal is not a perfect system built from scratch but a system that can evolve continuously without disrupting the customers who depend on it every day.

Making AI stick at the frontline
The statistic about abandoned AI projects points to an organisational problem as much as a technical one. Pilots that do not connect to measurable business outcomes eventually lose sponsorship from senior leadership, who see costs accumulating without a clear line back to business value. Tools that frontline staff find confusing or irrelevant get worked around and forgotten.

Banks need to close the distance between AI investment and business accountability. That means defining what success looks like in cost or risk terms before deployment, not after. It means involving the people who will use the tools in the design process, and it means treating change management as a core part of the programme instead of an afterthought.

The banks that are winning with AI are also investing in their people. They are helping staff understand how to work alongside these systems, where to trust the output and where to apply judgement. That kind of capability building does not show up in a technology budget, but it may be the most important investment of all.

The banks making meaningful progress with AI are not necessarily the ones spending the most or moving the fastest. They are the ones that have been deliberate about where AI fits into the business, honest about their constraints, and patient enough to build the organisational conditions for adoption to take hold.

Southeast Asia has the infrastructure, the consumer base, and increasingly the regulatory environment to be a genuine leader in financial services AI. What the region's banks need now is less focus on the pilot and more focus on what comes after it.