When Beatriz Martin began her banking career at UBS, traders still logged their bets on paper slips called blotters. They scribbled down each buy and sell, then handed the slips to a clerk who typed the entries into a system. Then came Excel, trading platforms and electronic execution. With each leap, sceptics prophesied job cuts. That fear proved overdone, says Ms Martin, who is now group chief operating officer and president for Europe, the Middle East and Africa at UBS: “Banking employs more people today than it ever did.” The work of bankers simply moved up a notch—from logging trades to making sense of data, thinking more about strategy and spending more time with clients.
This history is instructive amid the scramble to put artificial intelligence (AI) to work. Like most large companies, UBS is busily rolling out AI tools, training AI champions and measuring use. By June 2026, it had logged some 500 live use cases across the company. Yet amassing pilots is the easy part. The real headache for Ms Martin is figuring out what banking will look like as AI matures and what new things a 160-year-old institution can do with the technology.
Prepare for surprises
Banks are not natural habitats for experiments with new technology. Their highly regulated environment often prizes caution over curiosity. But the potential of AI is too vast to ignore, and UBS understands that the cleverest applications of the technology will bubble up from staff experimenting with it close to their work.
This is why the bank has set up a hub-and-spoke model, says Ms Martin. The hub is a central AI Office—staffed by technology, operations and risk teams—that sets standards for responsible AI use and brokers external partnerships. The spokes are within the bank’s separate business divisions and group functions, each with a dedicated AI head who oversees delivery and drives experimentation. Binding it all together are more than 1,000 AI champions and a network of regional “AI factories” that pressure-test ideas and keep ambition tethered to the bank’s risk appetite. (Read more in this series about how to build a dedicated “lab” to capture and scale grassroots innovation with AI.)
That architecture helps managers decide what to scale—and spot when a solution built for one purpose can also serve another. Ms Martin likens it to the utility of a kitchen knife. “Some will want to use the knife to cut meat; others will want to smear butter with it,” she says.
Take know-your-customer checks, the essential but often time-consuming work of verifying who a client really is. UBS applied AI to streamline the process—filtering out, for example, false-positive negative-news alerts that web searches throw up when several people share a name. But another team at the bank soon realised that this type of AI can also identify prospective clients.
Such discoveries are rarely pure luck. They rely, says Ms Martin, on a company’s culture. And in the case of AI, it is not enough simply to know what the technology can do; staff need the confidence to experiment, and the wit to spot how a fix in one department might solve a problem in another. UBS is trying to strengthen that habit through education and communication. The bank has partnered with Saïd Business School, part of the University of Oxford, first dispatching its top 250 or so most executives there to study how large organisations can deploy AI effectively. Now it is taking the academics on a global tour, with stops planned for London, New York, Zurich, Singapore and Hong Kong.
Know thyself
Philosophers still disagree over what the ancient Greeks meant when they carved “know thyself” into the stone wall of a temple in Delphi. One reading of the inscription proves useful boardroom advice: in the age of AI, corporate self-knowledge yields uncommon returns. If a company holds a competitive edge, such as clean data, AI often compounds it. Yet if a business is riddled with bad habits, algorithms will merely accelerate the mess. Taking a cue from the ancients, UBS approaches AI with two honest admissions.
The first is about what is broken. One example, says Ms Martin, is the way some processes have piled up at UBS after 160 years of trading, mergers and acquisitions. The result is a thicket of work-arounds, to which applying AI would be largely wasted. This is why UBS has built a tool called Opus, which inspects processes, proposes leaner ways of carrying them out and identifies opportunities to automate parts of them with AI over time.
To keep this simplification in line with its business priorities, UBS has drawn up nine strategic priorities at board level, which it calls “Big Rocks”. They range from client life-cycle management to the future of software engineering. Across each Big Rock area, UBS applies a common discipline: first, make the work visible by mapping the legacy process; next, remove steps that should not exist at all; and only then sketch how the process ought to run in a world where AI is increasingly capable and cost-effective. (Read more about decomposing roles into tasks to identify where AI adds genuine value in this series.)
This is particularly relevant as UBS continues to decommission legacy infrastructure inherited from Credit Suisse, the rival lender it took over in 2023. By June 2026, it had switched off 76,000 servers and retired about 1,700 business applications—roughly 60% of those targeted. Stripping out unnecessary complexity before adding AI is essential, says Ms Martin.
The second admission is about its edge. One unexpected advantage that UBS has found is in its mainframe systems—the hulking machines that still handle the heavy lifting of payments and ledgers in most big banks today. Decades of regulatory rigour have endowed them with an unusual degree of documentation and auditability—amounting to a supervised “fact base” that generative AI thrives on. UBS uses AI agents that can safely read and, in controlled ways, rewrite code.
Banking on bots
The breakneck pace of AI development has left many bosses unsure where to place their bets. UBS has chosen to be pragmatic. It looks past the much-debated route to artificial general intelligence (AGI)—systems capable of outperforming most people at most desk jobs—and considers how AI agents that increasingly perform complex, multi-step tasks can boost what the bank does.
Some of those agents are already at work. UBS divides them, broadly, into “makers” and “checkers”: the first type drafts the work and the latter scrutinises it, with humans on hand to adjudicate. As models improve and the cost of running them falls, UBS expects that this pairing will absorb more of the repetitive internal tasks, from rewriting legacy code and trawling through documents to reconciling scattered data. In Ms Martin’s telling, the wider use of agents will give staff more time for tasks that require greater judgement and relationship-building. (Read more about designing human-AI teams that outperform either alone in this series.)
The difficult task, as the number of agents at work grows, will be keeping order. Thousands of agents could one day run simultaneously, too many for any human manager to supervise unless the bank builds an “agent backbone” and a marketplace that encourages teams to reuse proven tools rather than to create a sprawl of untested ones.
What might this make possible? Consider a client who wishes to shift from a portfolio of individual investments into a discretionary mandate, giving UBS discretion to manage the client’s money. Ms Martin says an AI tool could help match a client’s holdings to the strategy’s underlying assets, finding the best overlap, analysing the tax consequences of switching and helping to minimise frictional costs. In most cases, a human adviser would still review and deliver the recommendation. But she sketches a future in which a client opens UBS’s banking app, asks the same question about their wealth and receives a tailored answer in seconds, ready to discuss with their adviser.
The technology is largely there. What is missing, Ms Martin notes, is clearer regulation on when and how AI may support investment recommendations. UBS is building the architecture so that humans and agents play distinct roles when the moment comes—and so that no part of the bank is caught flat-footed when it does.
Fostering AI-driven innovation
Cast your mind back to the paper blotter. Each previous technological wave ended up boosting what UBS does and Ms Martin is confident that AI will follow the same script because banking is a “people business”. Her focus is on building the right organisational architecture to capture the opportunities and surprises of AI as they arrive.
This approach anchors our Enterprise AI in action series. In another article, we examine how Rolls-Royce stress-tests its AI projects like jet engines to ensure they deliver clear value in exacting engineering work. We also draw on a conversation with Ethan Mollick, a leading thinker on AI in business, who argues that three elements—an empowered crowd of employees, clear leadership and a dedicated AI lab—are the surest way for companies to foster AI-led innovation.
This article is based on an interview with Beatriz “Bea” Martin, group chief operating officer and president for Europe, Middle East and Africa, UBS.
