2 days Ago By Lauren Towner
Trembling with excitement or paralysed by fear? We asked three experts how legacy banks should integrate a technology that’s ‘as powerful as electricity’ in its ability to transform organisations Since ChatGPT’s arrival two years ago it seems the world has been on tenterhooks, awaiting a generative AI (genAI) revolution. After false dawns with other technologies, the consensus is that this time it’s different. But what does genAI mean for banks today, and where is it taking them? We asked three industry leaders for their insight – Ranil Boteju, chief data and analytics officer at Lloyds , ING ’s head of advanced analytics strategy, Marco Li Mandri, and chief executive of software solutions company BBD , Kevin Staples.
The Fintech Magazine: We had tech consultancy Gartner tell us genAI was at the top of its hype cycle – it certainly seems to be entering a point of maturity. Will AI revolutionise banking? Is this time different? Ranil Boteju: Yes, it’s not hype. We’ve seen technology hype cycles before with distributed ledger, blockchain, Web3, metaverse, quantum.
But let’s be honest, none has made a huge impact yet, they’re very much top-down driven. Generative AI was different, within three months of ChatGPT coming out we had engineers all over Lloyds using large language models to solve real problems off their own bat. I’ve never seen that – that organic growth of use cases popping up.
I’m convinced this is a general-purpose technology that will transform organisations, like electricity did when it was discovered. Marco Li Mandri: I agree it will be transformative. But a common misconception is that this is easy.
Building this system takes quite some craft because the more you go into a core system, the more monitoring you have to do. Plus, you have to make the system safe, which is demanding. “I’m convinced this is a general-purpose technology that will transform organisations” Ranil Boteju, Lloyds Bank Kevin Staples : If you consider the sheer rate these generative AI models are advancing, it can seem terrifying.
And that leaves an organisation not knowing how to move forward. There is an inertia and it’s often related to the risk elements. My preferred approach with a technology that’s inherently not definitive, like this, where the outcome isn’t guaranteed, is a proof-of-concept approach where we adopt a mindset of let’s fail fast.
You’ve got to understand that because this technology is changing so fast, anything you adopt today will not be as appropriate as the better tech that arrives tomorrow. So, my advice is: keep it simple, look at use cases that will have value, then dive in, solve the problem, and realise quickly if you’re failing. The Fintech Magazine: So what is genAI being used for right now, and what is coming soon? Kevin Staples: There are three categories where it’s being adopted by banks and other corporates.
First is staff – we’re using these tools already in our day-to-day jobs as efficiency tools. Second is back-office processing – for our banking clients we do a lot of process re-engineering work and that’s a use case that’s within reach, it’s relatively straightforward to get there. And third is the holy grail – customer-facing applications.
I believe that, in the future, most banking customers will have their own ‘digital twin’ of a personal banker and this is an area I’ve been working on with our clients to achieve. It’s hard and we’re not quite there yet, but we’re overcoming a myriad of technical and business-risk complexities to head in that direction. Marco Li Mandri: At ING we’re focussing on several things.
First is scale – we identify use cases that can scale across all the markets and countries where we’re operating. The second one is about domain. In our case, there’s a lot of potential in the contact centre space, and also marketing personalisation.
GenAI can also help support analysts with their work in software engineering, and lastly, we’re using it in wholesale banking where we have a commitment to sustainability. “Keep it simple, look at use cases that will have value, then dive in, solve the problem, and realise quickly if you’re failing” Kevin Staples, BBD Ranil Boteju: We’re funnelling our use cases into four categories at Lloyds. For engineering, as Marco and Kevin touched upon – there are lots of opportunities for engineers to use coding copilots, automate test cases and documentation, and create synthetic data.
The second category is using genAI to create tools for frontline colleagues – things like call summarisers, post-call quality assessments – very important when you’re selling financial services products – and training tools. They’re working well. Third is tools focussed on the back office.
Then fourth, and the one we’ve paused on until we’ve got the right guardrails in place, is exposing generative capabilities directly to customers. The Fintech Magazine: What are the obstacles to scaling this technology?What is your own strategy? Marco Li Mandri: A few years ago, the challenge to scaling AI was around data. Now it’s about building a platform that’s embedded in processes that are core for the bank.
It takes time, but at least it’s clear what we need to do. On the technical side, you need a good roadmap so you’re clear about which use cases to focus on. You need scalable monitoring for when there’s a lot of interaction, and you need data-driven quality assurance.
On the people side, we know we need to educate all employees about the risks and opportunities that genAI can bring, and not just the people using it. And we see an increased need for what we call AI product leaders who will shape this technology. They know a particular banking domain, they understand IT, analytics, they care about the craft and so on.
But it’s a tough role. Ranil Boteju: AI, data science – these things have been around for more than 20 years at banks. GenAI posed additional risks, such as hallucination, misinformation, misalignment, toxicity and so on.
We’ve worked through manual and automated controls you can use. For hallucination, you can train the model on a more precise data set. You do something called ‘red teaming’ where people pose questions and if the answers are wrong, correct them.
Then there’s ‘ground truthing’ where you compare answers against a known database of answers. But the one we use mainly is having a human in the loop until we’re confident the answers are correct. So, it’s all about understanding the risks and putting mitigants in place.
Secondly, you need a platform that allows you to scale consistently. When we started, we had lots of people doing use cases that weren’t scalable, there was duplication, they were building on non-target platforms. So we’ve started to build out a clear target architecture with the capabilities we need to scale.
We’re building on our Cloud platform. Unfortunately, these things take a while! “A few years ago, the challenge to scaling AI was around data. Now it’s about building a platform that’s embedded in processes that are core for the bank “ Marco Li Mandri, ING Kevin Staples: I’ll repeat what Marco said earlier – it’s not easy.
Such a broad range of skills are needed to use these large language models (LLMs). Yes, you do need data science but you also need data engineering, Cloud platform engineering, you need machine learning ops and software engineering. It’s a hard, especially in an existing bank, to implement LLMs into production, and where business leaders need to be accountable for the money they’re spending but there’s no guaranteed outcome.
Which is why my go-to approach is find the right skills and partners, understand the need to fail fast, to realise value fast, chop and change as time goes on. It’s going to be a bumpy ride adopting AI, but quite an exhilarating one. The Fintech Magazine: How will genAI shape the next big shift in consumer banking? Marco Li Mandri: I always remind our teams about the 1970s and 80s when we put a computer on everyone’s desk and assumed productivity would increase.
It didn’t, because for a long time we just put a computer in every process, and the processes didn’t change. We must try to do the opposite with AI. Start from the domain and ask how you can re-engineer the entire process.
At ING, we link it to outputs, such as measuring straight-through-processing rates. For customers, decisions, such as for loans, will happen faster and hyper-personalisation will be a big shift. We also have an ambition to steer customers towards sustainability which we can do increasingly well, thanks to AI.
Ranil Boteju: Banking has been around for centuries but we’ve basically layered on new technologies. We haven’t reimagined what a bank should be. I expect people to ask what role a bank should play in people’s lives.
And with AI as a core tool, what could that look like? People will learn from the mistakes of digital and mobile where everyone just digitised their paper-based processes. We’re going to see organisations start from scratch, and it’ll be interesting to see who survives that, and who doesn’t. This article was published in The Fintech Magazine Issue 33, Page 34-35.
