What agentic AI means for financial services

Insights

• Agentic AI is here: It is moving from theory to real-world use in finance — helping teams, improving customer outcomes, managing risks, and driving growth.
• Types of agentic ecosystems: Two main types exist — personal agents that cater to individual needs and enterprise agents that optimize business processes and ensure compliance.
• Benefits: Agentic AI handles complex tasks independently, reducing manual intervention and enhancing agility and precision in financial operations.
• Challenges: Adoption brings hurdles like data privacy, transparency, legacy tech, ethical issues, and talent shortages.
• Recommendations: Financial institutions should rethink their operating models, invest in change management, focus on high-value use cases, and ensure scalability and oversight.

For financial services leaders, this feels like déjà vu. Just as nearly a quarter of financial leaders believed they had locked down a solid enterprise AI strategy and were ready to move forward, the next wave arrives — agentic AI. Executives must stay on top of the innovation cycle and turn these disruptions into opportunities.

Agentic AI is beginning to influence how financial institutions and fintechs operate, shifting from theory to early real-world applications. Deloitte predicts that by 2027, 50% of enterprises using generative AI will deploy agentic AI, up from 25% this year. Financial institutions are exploring the technology’s use to support teams across business lines and functions, improve outcomes for customers and shareholders, manage risks, and support long-term growth.

How financial institutions adopt agentic AI

The global agentic AI market in financial services is likely to grow from $2.1 billion in 2024 to $81 billion by 2034. Research shows that North American financial institutions lead adoption, investing in automation and intelligent decision-making capabilities.

Infosys sees two types of agentic AI ecosystems evolving in financial services: Personal agents that cater to individual needs autonomously, and enterprise agents that help financial institutions derive value from autonomous capabilities (Figure 1).

1. Personal agentic AI ecosystem

Personal AI agents are fundamentally changing how customers interact for their banking needs. These aren’t just digital assistants, they’re intelligent partners capable of more than simply answering queries or scheduling payments. Designed to replicate the actions a person might take in managing finances, personal agents go beyond providing recommendations —they autonomously act on those recommendations.

For example, when choosing a mortgage, a market evaluator agent sifts through offerings from a range of banks, weighing interest rates, loan durations, and customer service records, before presenting a prioritized shortlist tailored to an individual’s needs. A financial health analyzer agent reviews the applicant’s credit history, current financial standing, outstanding debts, and even spending habits to determine a realistic monthly mortgage limit. A reviewer agent evaluates the options and refines the shortlist. Once the customer selects a preferred option, yet another agent submits the application directly to the chosen bank, reducing friction in the mortgage application process. This agentic AI ecosystem delivers personalized service adapting to an individual’s risk profile in the most efficient way.

Fintechs are also developing personal agentic solutions that help customers find the best mortgage or credit card and handle the entire application process digitally. For example, Bud Financial, a fintech that supports financial institutions with data analytics, has rolled out a personal agent that helps low-income banking customers save on overdraft fee, averaging $460 per customer, with some saving thousands in unnecessary charges.

2. Enterprise agentic AI ecosystem

Large financial institutions are ramping up their efforts as well. Enterprise agents are being deployed to optimize core business processes, boost efficiency, and deliver scalable solutions organization-wide — all while ensuring adherence to regulatory and compliance standards. Mastercard’s Agent Pay lets trusted AI agents make secure, AI-driven international transactions. It uses tokenization for fraud protection that replaces sensitive card details with a unique, randomly generated digital token. Customers get more control through explicit permissions and purchase limits. Gartner predicts that by 2028, 33% of enterprise software will feature agentic AI, up from less than 1% in 2024, enabling 15% of day-to-day work decisions to be made autonomously.

Figure 1. Agentic AI ecosystems

Source: Infosys Knowledge Institute

Figure 2 shows how an agentic AI ecosystem fits within commercial banks to sharpen their prediction of payment defaults among micro, small, and medium-sized enterprises (MSMEs). The system deploys a series of specialized agents: An internal risk evaluator agent scrutinizes internal banking data for early signs of distress, a financial health analyzer agent analyzes external financial reports, and a sentiment monitoring agent monitors news and industry developments for red flags. Each agent assigns a weighted risk score — internal data carries the most weight, followed by financial health and sentiment analysis. A risk aggregator agent aggregates these scores, flags high-risk cases, and coordinates the process. Overseeing the entire process is an orchestrator, which manages the sequence of analysis and ensures a seamless flow of information between agents. By combining internal records, market intelligence, and real-time sentiment, banks gain a dynamic, comprehensive view of MSME credit risk — enabling quicker, more proactive decisions.

Source: Infosys Knowledge Institute

In a fully realized enterprise agentic AI ecosystem, most business processes will be powered by autonomous agentic workflows. Only the critical core processing tasks would be done in the traditional way. For example, in the cards business, all processes — starting from originations, customer servicing, dispute resolution, fraud management, and collections — would be driven by autonomous agents (Figure 3). These agents will make decisions on the fly, adapt to new information, and require minimal human oversight. But the critical tasks of accounts receivable (AR) processing and authorization will still be handled by the card processing engine.

Source: Infosys Knowledge Institute

What makes agentic AI different from traditional AI

Agentic AI stands apart from traditional AI in several ways. While traditional AI typically executes predefined tasks within static, rule-based workflows, agentic AI is designed to make autonomous decisions aligned with specific business goals. It operates with dynamic, adaptive workflows and continuously learns from user interactions and outcomes, rather than being limited to its initial programming. This enables the technology to go beyond the content creation or natural language processing capabilities of generative AI. Consequently, agentic AI can deliver hyperpersonalized, context-aware experiences for customers, moving well beyond the generic recommendations of conventional AI systems or the productivity gains and operational efficiency improvements of generative AI.

Source: Infosys Knowledge Institute

Operationally, agentic AI reduces the need for manual intervention — human oversight remains important, but these agents can handle complex tasks independently. Architecturally, we move from a single-model approach to orchestrating multiple agents and models that work together. Instead of a simple linear process, agentic AI follows an iterative cycle: It senses the environment, plans, acts, and then reasons from the results (Figure 4). This tiered approach is built on large language models as the core decision engine while deploying guardrails such as business rules and compliance checks to maintain control over deployment. This allows financial institutions to act on a broader range of business objectives with agility and precision. In wealth management, for example, agentic AI doesn’t just rebalance portfolios on a schedule — it analyzes real-time market data, simulates scenarios, and makes autonomous trades, delivering more responsive and tailored investment strategies than traditional static models.

Challenges in agentic AI adoption

While agentic AI offers a clear departure from traditional AI — delivering greater autonomy, adaptability, and personalized outcomes — it also introduces a new set of complexities for financial institutions. As banks and enterprises move beyond static, rules-based systems to deploy autonomous agents, they encounter challenges that extend well beyond technical integration and demand careful navigation to realize the potential of agentic AI.

1. Data privacy and security: Agentic systems require extensive access to sensitive financial and customer data, amplifying risks of breaches and unauthorized access. This autonomy heightens compliance challenges with regulations such as the EU’s General Data Protection Regulation and California’s Consumer Privacy Act. For example, a marketing or analytics agent retrieving customer data without explicit consent could violate privacy laws.
2. Transparency and explainability: Financial institutions operate in tightly regulated environments, where decisions such as loan approvals or denials must be auditable. The black-box nature of agentic AI complicates explaining rationales for outcomes, potentially masking biases or errors. This opacity hinders regulatory compliance and increases litigation risks — such as when mortgage applications are declined without clear justification.
3. Legacy technology and siloed data impede scalability: Enterprise deployment of agentic AI faces data fragmentation across siloed business units, preventing unified customer views and undermining accuracy. Legacy mainframe systems further impede integration with modern AI frameworks, creating operational bottlenecks. For instance, mortgage-focused agents cannot leverage core banking data, limiting cross-functional synergy. High-performance computing and scalable infrastructure are prerequisites, often lacking in traditional environments.
4. Ethical and regulatory issues: Historical data biases can perpetuate discrimination in credit scoring or investment advice. Accountability gaps emerge when autonomous agents make consequential decisions, making it difficult for regulators to assign liability for AI-driven errors. Global institutions face additional complexity adapting to evolving regional regulations, such as dynamic compliance requirements across jurisdictions. For instance, an agentic system designed to provide personalized investment advice could inadvertently steer certain groups toward lower-value products. Such outcomes risk perpetuating social biases and attracting scrutiny from regulators concerned with fairness and equitable treatment.
5. Lack of talent and change management: A shortage of AI-specialized talent hinders development, while legacy workflows resist integration with autonomous systems. Successful deployment demands coordinated change management across IT, operations, and compliance teams — addressing data readiness, continuous integration and continuous deployment (CI/CD) software development, and cross-departmental collaboration. Without strategic upskilling, employee resistance can derail implementation.

How financial institutions can adopt agentic AI

1. Rethink operating models: Break down silos between business and technology teams and ensure alignment on shared goals. This means integrating business and technology planning for agile development and deployment and enabling cross-functional collaboration among business, IT, compliance, and risk teams.
2. Change management: Adopting agentic AI is not just a technology upgrade, it’s a cultural shift. Leadership must champion experimentation, invest in upskilling, and establish strong governance to prioritize AI responsibly. The focus should be on learning fast, iterating, and keeping customer trust at the center.

Beyond strategy, the following recommendations, grounded in real-world implementations, provide a practical roadmap spanning strategic alignment, technical architecture, and responsible scaling for agentic AI adoption.

1. Define strategic objectives: Begin by clearly aligning agentic AI initiatives with your institution’s overarching business goals, whether those include improving efficiency, driving innovation, enhancing customer experience, or strengthening risk management.
2. Assess readiness and build foundations: Evaluate your current data infrastructure to ensure that data is clean, integrated, and easily accessible across the institution. Establish robust data governance practices that prioritize quality, security, privacy, and regulatory compliance. In parallel, develop comprehensive AI governance frameworks that include ethical guidelines, clear accountability structures, and mechanisms for transparency.
3. Start with high-value, lower-risk use cases: Target internal operations for your initial agentic AI deployments, focusing on back-office functions, IT operations, and data quality management where the risk to customers is minimal. Begin with controlled pilots or proof-of-concept projects to validate both the business value and technical feasibility before considering broader implementation.
4. Architect for technical scalability and interoperability: Adopt a flexible architecture that supports multiple AI models, providers, and agents, ensuring they can work together seamlessly. Leverage modular frameworks and platforms that facilitate multi-agent collaboration, effective memory management, and smooth integration with existing enterprise systems to future-proof your AI investments.
5. Implement phased rollout with human oversight: Start with a supervised autonomy approach, where humans remain in the loop for critical decisions and clear escalation protocols are in place. As your confidence in agentic AI grows, gradually increase the level of agent autonomy, allowing them to take on more complex and sensitive tasks over time.
6. Invest in explainability and continuous monitoring: Deploy explainable AI models that can clearly justify their decisions, essential for regulatory compliance and building stakeholder trust. Continuously monitor the performance of your AI agents, tracking outcomes, detecting biases, and conducting regular audits to ensure reliability and adherence to compliance standards.
7. Upskill workforce and manage change: Reskill and upskill your employees so they can effectively collaborate with agentic AI, with a focus on critical thinking, ethical oversight, and new roles such as AI trainers or governance specialists. Foster a culture that embraces AI, encourages experimentation, supports continuous learning, and proactively addresses concerns around job displacement.
8. Strengthen security and responsible AI practices: Implement strong security controls, including data encryption, access management, and regular vulnerability assessments to protect sensitive information. Embed responsible AI principles into every stage of the AI life cycle, prioritizing fairness, bias mitigation, and transparency to ensure ethical and trustworthy AI deployment.
9. Scale and optimize: Once initial pilots have proven successful, expand your agentic AI solutions across additional business units and processes. Continuously refine your models and workflows by leveraging feedback, ongoing monitoring, and regular audits, ensuring agent performance improves and adapts to evolving business needs.

Ultimately, the transformational potential of agentic AI depends on how quickly you adapt and refine your AI strategy to realize its full promise.

Well thought out execution to stay relevant

Agentic AI is not just an incremental improvement, it’s a strategic necessity for financial institutions and fintechs looking to stay competitive in a rapidly evolving landscape. It is not just a theoretical promise but is already delivering real results: From cost savings and operational efficiency to new product innovation and improved customer outcomes. That said, organizations that lead in adopting agentic AI, execute it effectively, and uphold responsible practices will stay relevant.