Financial services firms are under pressure to explore agentic AI, but many still face a gap between what the technology promises and what works inside regulated operations. Claims teams, underwriters, loan advisors and compliance functions deal with exceptions, judgment calls, legacy systems and regulatory nuance every day, which means AI adoption cannot be judged on proof-of-concept results alone.
Kieran Watts, industry lead, claims innovation at Digital Workforce, believes firms need to start with operational problems rather than technology-led pilots. He argues that the real test is not whether an AI agent can complete part of a process, but whether it can operate reliably in production, support human teams and leave a clear audit trail.
In this week’s Five Minutes With…, Fintechly speaks to Watts about why agentic AI is different from traditional automation, the risk of fragmented pilots, the importance of workforce adoption and why governance, second-line compliance and build-versus-partner decisions are set to move up the agenda.
Kieran, can you tell us about yourself and what brought you to this point in your career?
I have spent 30 years working in financial services operations across claims, risk management, compliance, and P&L leadership in global markets. That experience gave me a clear view of the challenges within the financial services and regulated industries, as well as insight into why many attempts to address these issues have fallen short.
My interest in agentic AI, specifically in financial services, and my work developing practical solutions for client use cases at Digital Workforce stem from observing the operational capabilities of this technology. It can perform substantial tasks independently, without needing constant human oversight, which is crucial in an industry that relies on experienced professionals to make complex decisions under pressure. Agentic AI has the ability to transform our industry’s traditional operating models, unlike any other innovation we have seen. Bridging that gap between emerging Agentic capability and practical business pain points is what drew me to this work.
What problem or opportunity are you most focused on right now?
My focus is to close the gap between what agentic AI promises and what survives contact with operational realities: a real underwriter, a legacy system, a compliance edge case or a regulator. Many programmes I have observed start from the wrong place. They begin with the technology and work backwards to a justification, rather than designing an AI programme with the operational problem as the focus. The temptation for organisations is to optimise for the 60% that looks good and leave operations teams to absorb the difficult 40%, including the exceptions, judgment calls, regulatory nuances, and workflow complexity that drive cost and risk. That model often results in fragmented AI initiatives scattered across a business, with no coherent adoption strategy.
There is also a significant blind spot around the total cost of ownership. Organisations tend to focus their analysis on the initial build and implementation costs, comparing what it would cost to build in-house against working with a partner, without accounting for what happens after deployment. Keeping an agentic system performing reliably in production requires specialist ongoing monitoring, maintenance, infrastructure and expert resourcing. That operational cost over the full life cycle of an agentic system can be far greater than the cost of building it. Organisations that miss this end up with costs locked in regardless of the outcomes, whereas the smarter approach is to defer ongoing Agent management costs to a specialist AI partner.
What do you think deserves more attention than it is getting in your part of the industry?
The people question. Minimal attention has been given to the workforce who need to operate alongside agentic AI every day. The claims teams, loan advisors and compliance teams whose cooperation is essential to making any of this work are also the people most likely to feel uncertain about what it means for them.
Take a loan application or a claim assessment. Initial intake, eligibility checks, documentation review, customer correspondence and routine status updates can all be handled by an agent working continuously in the background. The human stays responsible for the major decisions and complex commercial judgments. When people can see where their expertise remains essential and where the agent is lifting the administrative burden that was consuming their day, the conversation changes. People become eager ambassadors for the change, which in turn accelerates adoption.
Governance also deserves more serious attention. Second-line compliance in regulated industries often operates on random sampling, typically a small fraction of total volume, due to a limit on specialist knowledge and resources. Agentic AI removes that constraint by enabling the compliance function to scrutinise every transaction, every contract at scale, giving it a reach it has never had before. Agentic-enabled governance frees up resources to enable a richer, more dynamic and more insightful view of the enterprise risk landscape for Board-level oversight.
What do you think people still misunderstand about your part of the industry?
The most persistent misunderstanding is treating agentic AI as a faster version of existing automation in financial services. Earlier technology needed direction at every step. Agentic AI is different; it has the ability to work autonomously, to reason, to learn and to know when to escalate to a human in the loop. It can take a transaction through multiple stages, manage customer interaction, identify what needs escalation and close what does not, without someone initiating each action. It opens up a different set of questions about how an operating model should be designed to achieve the desired outcome, rather than focusing on simply replicating existing processes.
The second is believing that running several pilots adds up to a strategy. Without an overarching, clear roadmap connecting individual POC’s to enterprise outcomes, most pilots remain pilots and progress stalls.
There is also the assumption that risk aversion is a reasonable basis for waiting. Late adoption in this market risks becoming quickly uncompetitive, and the ground ceded to faster-moving competitors may never be recovered.
What do you expect to rise up the agenda over the next year?
Two key themes are emerging: the ‘build v partner’ question and ‘AI governance’.
The build versus buy versus hybrid dilemma will move firmly up the agenda as financial services firms move beyond proof of concept and face deployment decisions. The ongoing management and governance of Agents is becoming a highly specialised and complex challenge with the rapid development of the newest models. Organisations will need to decide if they want to make the material investment needed internally to manage that challenge or to focus on their core business and collaborate with an AI specialist partner to derisk that responsibility.
Governance and the proof of agentic output will become a much more important narrative. Regulators and boards will require full transparency, observability, clear explanations and audit trails as AI assumes greater decision-making roles. Organisations will need to demonstrate that their compliance processes have consistently worked at scale across every relevant transaction where an Agent has been empowered to make a decision or action.
What is the one question about your company we should have asked, and what is your answer?
As a company working at the forefront of Agentic AI in the insurance and financial sectors, the common question we are asked is, ‘Where is Digital Workforce seeing the biggest opportunity in regulated industries right now?’
The area where we have seen the most exciting potential of Agent capabilities is in second-line compliance, delegated authority and risk functions within insurance & financial services. In increasingly regulated environments, these teams carry significant board-level accountability but have always operated under the constraint that they can review only a small fraction of the total volume of transactions. If something goes undetected, the organisation only finds out when an auditor or regulator finds it first.
Agentic AI removes that constraint in a way that no previous technology has. It becomes possible to review every transaction at scale and frees up specialist resources to focus on high-value risks. Digital Workforce is focused on the high-stakes challenge of governance in regulated industries, where the cost of not knowing can be measured in regulatory fines and reputational damage.
