Agentic artificial intelligence (AI) presents exciting opportunities in the financial services industry, with the potential of transforming customer experiences, and delivering significant business value.

However, the technology also introduces significant challenges, including goal misalignment, misuse of tools and application programming interfaces (APIs), and data privacy concerns, according to a new report by IBM.

Agentic AI refers to AI systems that exhibit a degree of autonomy and goal-directed behavior. These systems function as agents capable of making decisions, taking actions, and pursuing objectives over time with minimal human intervention. They are able to solve complex problems, interpret and create actionable plans, and execute these plans using a suite of tools.

According to IBM, agentic AI holds transformative potential in the financial services industry, promising automation, personalization and streamlined operations across complex business processes including customer onboarding, know-your-customer (KYC) and anti-money laundering (AML), fraud detection, and risk assessment.

Customer engagement and personalization

The report present customer engagement and personalization as key areas of agentic AI implementation. Applications include hyper-personalization of product and service offerings, dynamic pricing, as well as tailored recommendation and robo-advice.

The report shares one example of agentic AI implementation in customer onboarding and KYC workflows.  In this process, a customer would by submitting an account application along with the necessary documents. A customer service representative would manage this interaction, assisted by the agentic system.

At the core of this agentic system, a principal agent would oversee the entire process, using an orchestration framework to determine capabilities and access. This principal agent would delegate tasks to domain-specific service agents, such as a risk analysis agent or a sanctions screening agent.

These service agents would then coordinate with task agents responsible for actions like document validation and ensuring AML/KYC compliance. If a case is high-risk or requires further verification, the agents would trigger human intervention.

Finally, the customer service representative would review and confirm the outcome to ensure accuracy and regulatory compliance.

Operational excellence and governance

The second use case outlined by IBM involves optimizing middle/back-office operations. In this area, agentic AI promises reduced risk, enhanced compliance, and streamlined workflows and administrative overhead, as well as enhanced business outcomes.

Emerging applications in this area include lending and loan approvals, account operations, anomaly detection, automated risk management and compliance, and business support operations.

Several organizations have already started integrating agentic AI in this domain. PwC, for example, has introduced AI agents capable of automating complex tax tasks such as handling K-1 forms, which report income from investments, Dom Megna, PwC’s US AI Tax Leader, told CFO.com in a recent interview.

These forms often arrive in messy formats with unclear information, so humans typically need to interpret and map them manually. Now, AI agents are trained to understand these documents, decide how to classify items like deductible expenses, and suggest actions with a confidence level, he explained.

In Germany, private bank Metzler recently partnered with Swiss software startup Unique. Unique provides an AI platform enabling financial institutions to deploy agentic AI into their back and middle-office functions through 25 off-the-shelf use cases. The startup, which serves the likes of Pictet, UBP, SIX, LGT, and Partners Group, secured a US$30 million Series A round in February to fuel its growth and global expansion plans, particularly into the US.

Technology and software development

The third application of agentic AI outlined in the IBM report is technology and software development, an area where these systems have the potential to significantly enhance operations, software development lifecycles, and infrastructure management.

Internal experiments at Infosys, an Indian multinational technology company, revealed that AI agents can achieve between 80% and 90% improvement in database code generation; between 60% and 70% improvement in generating application programming interfaces (APIs) and microservices; and up to 60% improvement in generating user interface code.

New risks and challenges

Despite its benefits, agentic AI presents unique risks that require careful oversight. These systems are inherently complex, making their behavior difficult to predict and manage, and require careful consideration and tailored risk management strategies.

Key risks outlined in the report include goal misalignment, where systems may pursue objectives that conflict with an organization’s true intentions or ethical standards. There are also concerns around safety, accountability, and regulatory compliance since agentic AI systems are able to act without constant human oversight.

Furthermore, since agentic AI combine tools and APIs in creative, unexpected ways, they may create vulnerability issues or cause operational failures. These agents might also gradually overstep their assigned responsibilities and begin making decisions or taking actions that should require human approval.

Finally, by autonomously accessing, processing, and retaining sensitive data across systems, agentic AI comes with heightened privacy risks. This includes the risk of unintentionally leaking personal or confidential information, especially if the system uses persistent memory or integrates with tools in ways that bypass standard data protections.

Adoption of agentic AI on the rise

Though adoption of agentic AI remains in its early stages, momentum is building. A recent industry survey by

Wolters Kluwer, a Dutch information services firm, found that 38% of finance leaders plan to adopt the technology in the next 12 months. By 2026, adoption is projected to reach 44%, representing a more than 600% increase. Currently, only 6% of finance leaders are employing agentic AI.

Featured image: Edited by Fintech News Switzerland, based on image by sodawhiskey via Freepik