Financial services providers are moving away from pilots and proofs of concepts towards scaled, operational deployment, according to panellists at Microsoft AI Tour London.
We are now experiencing the agentic wave of AI, according to Christian Sarafidis, financial services leader for Europe, the Middle East and Africa at Microsoft. Rather than viewing AI as a bolt-on productivity tool, he framed it as a catalyst for redesigning end-to-end processes: “Start by having your own agent as an assistant (for emails or translations), then move onto deploying task agents (for onboarding or checking docs) and ultimately think about end goals.”
That evolution reflects a broader inflection point. “There is a shift from experimentation to operation,” said Sarafidis, adding that in financial services in particular, “you must know what your return on investment is.” Leaders are now focused on prioritising investment and identifying where AI can deliver measurable business impact. “You all sit on tonnes of data, what makes an efficient AI deployment is translating the data into intelligence,” said Sarafidis.
Dave Collier, financial services lead for Microsoft UK, posed a question many enterprises are struggling with: how do organisations scale AI and move beyond experimentation?
For Simon Bullers, chief technology officer for the Bank of England, the answer has been ubiquity. AI has been “deployed to every colleague, despite what department they work in,” he said. “The goal is not isolated innovation teams, but organisation-wide enablement.”
The bank currently has 2,000 agents – and expects the number to rise to 5,000 – across 5,800 of its employees.
Emily Prince, group head of analytics and AI at London Stock Exchange Group, echoed the need for clarity of purpose. “Be clear on problem you’re solving,” she said. “As an industry, we’ve got very accustomed to build in segregation. What we’re seeing now is the breaking down of those processes and reinventing how we solve problems. Culturally, it’s gone into every segment of LSEG.”
Then, Will Hyams, director of AI productivity for insurance broker Howden Group Holdings, reinforced the universality of AI’s potential: “Every role in some way can benefit from AI. Helping everyone understand how you can help them is key because once individuals see personal value, they’ll be able to see how it benefits others.”
One of the most significant changes that AI has brought to financial services, according to Prince, is the democratisation of innovation. AI tooling, increasingly embedded into enterprise platforms, is reducing barriers to experimentation and deployment.
As AI systems become more autonomous, governance and trust remain front of mind – particularly in highly regulated industries.
Bullers described the Bank of England’s approach as grounded in “trusted AI”. “It’s a copilot, not an autopilot, there’s a human somewhere in the loop,” he said, underlining the continued role of oversight even as agents take on more tasks.
Meanwhile, Prince highlighted the complexity of deploying agentic AI in regulated capital markets. “We’re in a hard industry where the stakes are high,” she said. “At LSEG, we’re thinking about the marriage between operations and models that we have used for decades.”
Stress testing, for example, involves “very complex, different asset classes and conventions.” By bringing agentic AI workflows into the process, LSEG has reinvented something that used to take multiple steps for many years into a single workflow.
A recurring theme throughout the session was the shift from coding-centric AI development to domain-led innovation.
“You don’t need to be a coder,” said Hyams. “Historically, the challenge was how to get a cyber broker expert to work with a coder to do something. Now, organisations can take that broker and train them in AI to use their domain knowledge.”
Prince added that combining human expertise with AI agents is changing how data is used. The result is broader and more balanced data consumption. “People are now starting to use more rounded sets of data,” she said. “They are overcoming the huge bias to more recent data as more comprehensive historical datasets become accessible through AI.”
At the same time, cultural hesitation remains. Even with embedded copilots for email and collaboration tools, Prince noted she still hears questions from team members like “am I allowed to do this or is that cheating?” when using AI agents to support their workloads.
When Collier asked how productivity gains are measured and translated into tangible value, panellists cautioned against narrow metrics.
Prince highlighted that while many AI initiatives start as productivity plays, expectations quickly shift. “It started as productivity gain but we all came back to demand more,” she said. “Some of the most successful adopters are those running and executing [agents for jobs that] would have taken a long time. Now, they can be done in an afternoon.”
Hyams challenged the perception that AI is prohibitively costly.
“AI isn’t hugely expensive – two or three different use cases can pay for everyone’s licenses,” he said. “It’s not just about time saving, it’s about things we didn’t do before but now can do, such as deeper research into prospective clients.”
Across the discussion, one message was clear that the “agentic wave” is less about isolated copilots and more about re-engineering how work gets done.