Lloyds Banking Group's agentic strategy is often small, often deterministic and often supervised

Lloyds Banking Group has been in the midst of a business and technology transformation for many years now - which comes with the territory of ongoing change. Thecurrent program is called Accelerate 20230and builds on the previous work of dismantling legacy systems, moving to the cloud and aggressively hiring data and technology engineers. Accelerate is aiming for £2 billion of gross savings by 2030 and CEO Charles Nunn has said that the priority is to embed AI to make finances simplified and more personalized than ever. As such, it was with interest that I got the chance to sit down with Clare Schramm, Managing Director of Transaction Banking Technology Platforms at Lloyds, last week at an Acceldata conference in central London. A key takeaway after speaking with Schramm is that the bank is thinking about agentic AI with an aggressive curiosity: pointing it to use cases that make sense, whilst also recognizing that it’s early days, the benefits may be small to start, they may focus on deterministic workflows with supervision and they may not be glamorous. She said: > As much as AI is there and it is providing a lot of value, it's still very much in research and development mode, right? So we do have to operate in a very controlled way. Our customers trust us, so we operate with a lot of guardrails with AI. It’s worth stating that transaction banking is the corporate side of a bank's plumbing - the business that moves money for companies and institutions day to day, rather than lending to them or advising them on deals. It covers payments and clearing, cash management (helping a corporate see and control balances across lots of accounts), liquidity and FX, trade finance, and merchant services (processing card payments for businesses). It sits apart from retail banking, which is your current account and mortgage, and from investment banking, which is M&A and capital markets. Schramm has a strong view of how agentic AI should be used in payments - particularly around how institutions should work together on open standards for agentic payments. Schramm is also on the board of OpenUK andrecently wrote a blog advocating for collaboration to ensure open standards for agentic payments. She told diginomica: > I think the biggest thing is, in the future, we could imagine a lot of agents operating on our behalf. And if they don't have a common protocol, or some sort of common understanding amongst them, it risks the same problems we have as humans and as individuals if we don't have a common language we can operate in, right? She said that the open source community is the place to convene on the best standards, ones that the tech industry can agree and support. She added: > The last thing we want is a bunch of agents that don't all speak the same language trying to make payments on your behalf. That's when the whole agentic promise, that future, collapses. I do see it as quite existential. > > > I think the open source community has always been that place in tech where we drive the things we have in common - baseline things we can agree on, and good principles and standards for jurisdictions and geographies. Lloyds use cases Lloyds Banking Group’s CEO has said that he wants AI supporting every customer interaction. Schramm’s platforms may not be directly customer facing in the traditional sense (e.g. retail), but there are a lot of corporate and government payments being managed - in fact Schramm said that Lloyds is the largest sterling clearer in the world. She said that her and her team have been leaning a lot into AI. The bank uses a variety of technology partners including Microsoft Copilot, Claude Code from Anthropic, Google Gemini and Google Cloud for its data warehouse/lakehouse. The focus initially has been on help and support use cases, but has since extended to onboarding the government onto its systems (it won a big contract to process payments for government banking in August 2025) and then shows them how to use the systems. Schramm said: > They're really small use cases, but really, really high impact in terms of responding to demand and the increase in volume that's going through our systems. And Schramm pointed to an approach that is becoming increasingly common (and accepted as best practice across the industry): using generative AI as a reasoning engine or to problem solve a query, but then coupling this with deterministic workflows to ensure accuracy. She said: > Now we're starting to look at different parts of our payment journey. Looking at agents in payments - there are oftentimes a lot of errors in payments. Where can we use a deterministic workflow to proactively correct some payments before they go through? Simple things like address errors and address validation - correct that stuff, so that a human doesn't have to intervene, and they can spend their time on a higher-value use case. > > > We've seen some really good results so far. Again, we start small. We're being really deliberate about working alongside humans, and over time we're gradually increasing the usage. Another use case that has sprung up is Foreign Exchange (FX) - using AI to predict when the best rate is available for customers. Schramm said: > This is quite a big business for us, in terms of how we advise our customers on how to manage their money flows between sterling, euro, dollar, various different currencies. Again, because it's a deterministic workflow, and we have all this history about currency fluctuations and economic climate… > > > We've seen real bottom-line impact as a result of deploying some of these tools and arming our people who manage our customer relationships with these tools to advise customers better. > > > It will continue to evolve over time, and I imagine it will grow into potentially autonomous agents. Going slow However, despite the ambition for autonomy in certain areas, Schramm is also clear that Lloyds is starting small, including humans in the process, change and supervision, and is being incredibly cautious about where AI is introduced. In particular, she’s cautious about introducing AI that turns people away. She said: > You don't really slopify your very trusted banking application [with AI]. So I guess how we go about doing it is to look at it from a productivity lens - tips and tricks, productivity, right? > > > We've been living with chatbots for at least 15 years now. So we're building off of that, and that's where our early help and support use cases grew out of - like a chatbot, chatbot plus plus, right? They can do a lot more, they're more context-aware, because people are quite used to that. But I agree - you don't want to over-AI-ify. Schramm said the bank has been quite deliberate about not overdoing it and not overwhelming people. So far Lloyds has said that generative AI delivered £50 million in value in 2025 and it anticipates a further £100 million in 2026. A modest amount in the context of the multi-billion pound savings from the Accelerate programme as a whole. Whether that’s a reflection of the value being delivered by generative AI, or Lloyd’s cautious approach to adoption, it’s hard to say. But Schramm said: > [We’ve been] keeping that trust principle, and working within the boundaries that customers are used to. If they're used to the relationship, we still index on the relationship they have with a relationship manager, and just arm that relationship manager with some great AI tools. We work within existing help and support constructs that they're already comfortable with today. And on her final thoughts, the guiding principles should bring some comfort to Lloyds customers. Schramm said: > I think we still always need a human in the loop, especially when we're talking about money flows. But if it helps to speed along a process, I see that as goodness - as long as it's within established guardrails, people can audit it, and people can stop it from going haywire