Just about each firm faces the problem of remodeling itself within the AI period. Few must steadiness that with managing a multi-year integration of its greatest rival. However that’s precisely the state of affairs at UBS, which took over fellow Swiss financial institution Credit score Suisse in 2023.
Stephan Hug is UBS’ head of group capabilities expertise, a greater than 5,000-strong group which spans threat, finance, compliance, HR, authorized, comms and branding and company providers expertise. He got here to UBS after over 20 years at Credit score Suisse, together with eight years as its chief architect.
On this dialogue, edited for size and readability, he explains how the financial institution balances the dangers and advantages of integrating AI into its operations, who actually advantages from AI, and what digital and knowledge sovereignty means for the famously impartial European nation.
The place are you at together with your AI journey?
Stephan Hug: We’ve got been very busy within the final two and a half years with the combination of Credit score Suisse. That has been and nonetheless is our primary precedence. Thus far, we have now been extraordinarily profitable – nothing [has] failed. By the tip of 2026, we must be materially accomplished.
A few 12 months in the past, the group government board, below the lead of Sergio Ermotti, our CEO, stated, “Look, we are able to’t actually wait till this entire integration is over. The world is progressing. We’ve got to make an funding into AI.” And so, they got here up with one thing referred to as AI Imaginative and prescient 30, which is mostly a four-year pointed funding into AI throughout the agency. And that is available in many, many alternative shapes. We’ve got deployed Copilot to all our staff. That’s most likely what’s most seen to a number of our staff.
Then we began defining what we needed to do in AI. What have been the larger constructing blocks we need to put money into to mainly transfer UBS alongside to develop into an AI-enabled firm? There are a few very foundational components. One is clearly round governance and controls, as a result of we have now to get that proper. We’re a closely regulated financial institution. We’re a Swiss Financial institution Belief, and our popularity is essential.
How do you make sure that AI fashions don’t introduce unacceptable threat?
We initially had a platform referred to as Threat Lab, and because the title suggests, we use that to develop and check threat fashions very efficiently. [Enterprise AI firm] Domino Knowledge Lab is mainly underpinning Threat Lab. So, it’s an built-in platform the place we are able to develop, check, and finally deploy fashions. It’s built-in with a knowledge platform that runs on high of Databricks in my organisation. It has confirmed to be extraordinarily helpful. After which when AI got here alongside, we determined that we’d use Threat Lab throughout the agency and would lengthen it to AI fashions as properly.
Conventional deterministic threat fashions have been a part of banking for years. So the place does Gen AI slot in?
You may say the in a single day success of Gen AI and enormous language fashions is what triggered our government board to look into that and say, “Hey, if we don’t need to miss the practice, we have now to take a position,” and GenAI performs an vital position in our journey.
We do have an inside model of ChatGPT, referred to as Crimson, which supplies entry to ChatGPT in a safer trend than simply utilizing ChatGPT over the web. We are able to mainly management what knowledge will get despatched the place.
We’ve got seen a number of chatbots popping up throughout the organisation to resolve sure issues. For instance, as a financial institution, we have now a number of insurance policies. They’re normally very difficult. We’ve got written a chatbot that enables us to question the universe of insurance policies. So, I can say, “I’m travelling to New York, I need to invite my group to a dinner. What’s my allowance? What do I’ve to do?”
How do you handle the rollout of AI fashions within the organisation?
To start with, we’d like a list of all fashions. That’s very, crucial. That’s vital with the danger fashions [the bank uses]. That is equally vital for any machine studying mannequin or some other AI mannequin. We have to perceive what the use case is and what it’s getting used for. We have to perceive how the mannequin has been examined. We have to perceive the place the mannequin is deployed.
We did the identical with purposes. Is the appliance SOX related? I’ve to grasp who the proprietor is. I need to perceive the place it’s deployed. I need to perceive who makes use of it.
We’re a wealth supervisor, and that comes with sure values. We’re not going to permit, not less than within the first spherical, AI fashions that make daring selections round your investments, that make suggestions to you. Initially, we are going to at all times put the human within the loop.
However you possibly can mainly use the AI to give you concepts at scale that most likely a monetary advisor by himself couldn’t give you, as a result of we are able to analyse your behaviour. We are able to analyse your investments, after which say, ‘Oh, purchasers that did one thing comparable, or have an identical profile to you, did this. Would that be an concept for you?”
AI and knowledge sovereignty are massive speaking factors throughout Europe proper now. How are they shaping your AI technique?
For me, it’s extra about the place the information resides. We completely must guarantee that Swiss knowledge solely resides within the Swiss area of Microsoft Azure. Swiss banking legal guidelines, Swiss banking secrecy, don’t permit us to retailer shopper and client-identifying knowledge outdoors of Switzerland.
Switzerland is a really small nation. It’s solely 9 million individuals. Sure, it’s wealthy, however I don’t assume we have now the bandwidth to actually develop a sovereign cloud. Swisscom is making some strides towards that [goal], however they are going to be massively behind the three massive cloud suppliers. As a result of we’re a world organisation, we’d like a globally constant infrastructure.
Our two main tech universities, ETH and EPFL, introduced a Swiss LLM. We will certainly look into that. Will it utterly change what we do with OpenAI? No.
Leave a Reply