EXCLUSIVE: “Agents of Change” – Jon Lofthouse, Citi in ‘The Fintech Magazine

While Citi’s 2,000 human AI Champions are driving a new operational era at the bank, its Citi Stylus Workspaces platform is poised for an agentic upgrade

Jane Fraser’s tenure as Citi Chief Executive has coincided with a huge push to integrate AI across the global bank.

Almost all of the firm’s 240,000 staff can now call on AI tools to boost productivity, and in June, a group email encouraged them to ‘jump in and start exploring’. Citi spent almost $12billion on technology in total in 2024, but its the bank’s suite of AI tools that grabs the headlines. They include intelligent call handling systems, compliance and AML monitoring solutions, models for market movement prediction, and a platform that creates investment strategies for wealth management clients.

In Fraser’s own words, ‘the future is agentic’, and Citi has begun rolling out next-level systems that can suggest workflows for tasks, then carry them out. The CEO joined Citi in 2002 during its wild acquisitional phase ahead of the credit crunch. For years after the economic crash of 2008, the group was considered too big, too complex, and held back by dislocation caused by silo working. Then came a series of operational and  regulatory failings that compelled a rethink.

Alongside decommissioning legacy IT systems and removing a senior tier of management, AI has played a core part in Fraser’s turnaround of the business, which has this year enjoyed sustained share price growth after 16 years of flatlining. With AI technology cutting processing time, improving cash forecasting accuracy, and cutting false positive AML checks within some areas of wealth management, the success of these applications is driving adoption elsewhere. Citi’s Chief Information Officer, Jonathan Lofthouse, believes AI development is now surging.

“We’re in a new, exciting generative AI phase where we’re able to apply large language models (LLMs) to a whole new set of use cases,” he says. “We’re realistic about the speed that we should go to ensure systems are safe. Looking at how good the LLMs have become, though, who knows what’s going to be possible even one year from now?”

In June, a management restructure underlined Citi’s intent by putting its Head of Technology and Business Enablement and Chief Operating Officer in charge of a global AI rollout. Around 4,000 staff have been designated ‘AI Champions’, and accelerator training courses keep workers up to date. In October, around 175,000 staff were given training in how to write effective AI prompts to ensure they were making the most of the tools on offer. It’s both a top-down and a bottom-up strategy, while the tools themselves are deployed horizontally and vertically, says Lofthouse.

Covering all angles with AI

Horizontal tools are used across departments to boost productivity and cohesion among staff; vertical tools are for use cases tailored to specific clients and the teams that serve them. Among the horizontal tools is Citi Assist, a resource that finds answers to worker queries around the bank’s policies in areas such as human resources, risk, compliance and finance. Following feedback from users, it was subsequently integrated into the bank’s Microsoft Teams applications.

Another is Citi Stylus Workspaces, which can summarise data such as audio files or documents and is used across the business – from Fraser herself to coders and customer service staff. Lofthouse explains that Citi Stylus Workspaces is a retrieval-augmented generation platform, which is connected to an LLM with an external knowledge base. That means it can retrieve both information from within Citi’s own communication channels and file storage systems as well as from public-facing websites.

The Citi Stylus Workspaces genAI system connects to around 100 business applications and data repositories (such as Slack, Jira, Teams and GitHub) where it crawls and indexes the data to deliver solutions that fit Citi’s specific needs.

“Traditionally, Citi Stylus Workspaces would give you an answer using LLMs. You were able to upload documents and reason about those documents and ask questions,” Lofthouse says. “We then connected the Citi Stylus Workspaces platform to many of our internal systems – some simple things such as our global directory, and platforms such as Jira [the project management tool] and the data corpuses that we have within our engineering teams.

“We’re looking to integrate that platform with more systems, because the more integration we do, the more powerful Citi Stylus Workspaces becomes, and the more it’s going to help our colleagues make things run markedly faster.”

The bank recently revealed that Citi Stylus Workspaces now features an agent model, which Lofthouse says will be the ‘next big thing’.

“As well as being able to ask questions, you’ll be able to give that agent a higher task to do,” he adds. “It will show you a work plan for your task, then reason about the answer it gives you.

“Citi Stylus Workspaces is now a work planner that, once you’ve approved [the proposal] as an end user, it will then go away and complete an entire task.”

Some might suggest that this higher level of automation increases the potential for mistakes, but Lofthouse says that – at this stage – humans remain very much ‘in the loop’. He says: “We have a lot of governance internally, whether that’s around which processes we add AI to, the models we’re using in the underlying technology, or ensuring we’re thinking about things like cross-border data.

“Many of our tools will cite the underlying data used by Citi Assist, for example, which is restricted to a specifically curated data corpus that we’ve curated for that assistant and not public domain data from the internet.”

The vertical use case

In late August, Citi revealed the launch of two AI platforms for its US wealth management teams – AskWealth and Advisor Insights – with global rollouts planned for early next year.

AskWealth is a genAI-powered conversational assistant that provides service teams, advisers and managers with instant answers to client questions about their portfolios and the wider market. Advisor Insights is a tailored dashboard that updates Citi Wealth bankers and advisers with news and data about market movements, portfolios and events.

Both aim to free up staff members’ time, so they can focus on client relationships rather than simply gathering information. Elsewhere, Citi uses LLM technology to manage customer calls, guide colleagues, and analyse conversations  after a caller has hung up.

Lofthouse says instead of a caller navigating options via their keypad, the bank’s interactive voice response (IVR) system assesses, then delivers the information the caller requires, or, if necessary, puts them through to a customer service worker.

“Clients speak to our IVR platform and go straight to the piece that they need – which then reduces our call handling time and improves client experience,” he says. “When someone does need to speak to one of our agents, the LLM will continue listening to help our agents find the right pieces of our servicing platform to quickly drive them to the thing the client wants. And after the call, our agents get a call summary.

“We find the platform’s summary has better data quality, so we’re able to run better analytics over why our clients call us. It reduces the time our agents spend on administration so they can spend more time with our clients. We think this massively improves the client experience.”

Lofthouse refuses to be drawn on how AI in the wider world might develop. But he is optimistic about thetechnology’s ability to turbocharge the bank’s processes.

“The LLMs provided by the hyperscalers [Cloud-based infrastructureplatforms such as Google] are going to unleash a whole new world of possibilities for our use cases,” he says.

“The world is moving very fast, and we’ll continue to have guardrails and checks. But this is an exciting time when we think about the possibilities for modernising our systems.

“Whether that’s giving our clients chatbots, better voice-to-text capabilities on our IVR platforms, or better capabilities in terms of the information we deliver to clients, we think AI can be relevant in all of those spaces.”

This article was published in The Fintech Magazine Issue #37, Page 30-31

The post EXCLUSIVE: “Agents of Change” – Jon Lofthouse, Citi in ‘The Fintech Magazine appeared first on FF News | Fintech Finance.