The Evolution of AI in Financial Services

With the banking sector undergoing a profound transformation, Agentic AI is at the centre, driving a pivotal shift for the practitioners responsible for risk, quality, and delivery in the BFSI space. Deloitte’s State of Generative AI 2025 report reveals clear insights into how Indian organizations perceive and adopt AI: over 80% of Indian organizations are already actively experimenting with autonomous agents. Over the past decade, the BFSI sector primarily utilized strict, rule-based automation. Agentic AI is a major step forward. In the context of AI in financial services, agentic AI acts as goal-oriented problem solvers, enabling human practitioners to act as supervisors and handle exceptions. Agentic AI acts autonomously, while learning from each experience, adapting to take care of both routine and more complex scenarios without any handholding.  

It continues to play an important role in the QA function. With re-learning at its core, agentic AI learns from its experience in each test cycle. It is becoming increasingly skilled at identifying vulnerable areas, adapting to changes, and even recommending testing strategies based on past patterns. Yet, a lack of trust in AI-based outputs persists. With compliance at stake, off-the-shelf solutions still create concerns around accuracy, coverage, bias, and explainability. It is important to ask: Can these agents flag edge cases, boost coverage levels, and explain their decisions at the pace regulators now demand?  

In 2025, QA leaders face a big challenge: turning this agentic capability into outcomes that comply with the complex regulatory requirements and remain trustworthy to customers.

The Key Drivers of the Shift to Agentic AI

Compliance in 2025

The compliance landscape in the BFSI sector is getting more complex by the month. In 2025, regulators are expecting real-time assurance that ensures that decisions taken are explainable. Agentic AI’s real value lies in its ability to create systems that automatically monitor and adapt, while providing clear audit trails and a rationale behind every decision made. For QA, teams must be able to measure results clearly: better coverage, more effective defect identification during testing, fewer defects in production, and the confidence to release updates.

Demand for Personalisation in Financial Services

Customer expectations continue to change. Both banks and customers are strongly emphasizing personalization, as customers seek better digital experiences, while banks are realizing the positive impact personalization has on customer retention. Banks that focus on building and maintaining emotional connections see a notable increase in customer loyalty and lifetime value. For example, banks are now ensuring ready availability by operating 24/7/365 contact centres across various channels, including live chats. 70% of customers expect a consistent customer experience across channels. This is one of the major factors influencing a customer’s decision when it comes to choosing their primary bank. Agentic AI can build on this by learning what matters to each customer, enabling it to deliver personalized recommendations that are truly valuable across a range of scenarios. In fact, Gartner predicts that by 2029, Agentic AI will resolve 80% of common customer service issues autonomously.

Financial Crime

Mitigating financial crime is still a lucrative area for Agentic AI within AI in financial services. Banks use up to 10-15% of their staff on AML-related activities alone. Interestingly, between 2015 and 2022, the financial industry was able to detect only 2% of global financial crime flows, despite a 10% annual increase in their know-your-customer and anti-money-laundering activities every year during this period. Agents can collaborate with other agents independently through a multi-agent system for AML-related investigations. The subsequent agent acts based on the previous agent’s action, and the role of human practitioners lies in validating the entire report.

Adopting Agentic AI for BFSI

While powerful drivers towards the adoption of agentic AI exist, legacy systems and skill gaps still create risk pockets. However, considering the benefits agentic AI brings to the table, QA leaders need to find ways to integrate the same into their existing systems.  

A practical solution is to focus on incremental rollouts with regular quality checks. Teams can start by addressing a specific area or bottleneck rather than an end-to-end transformation from day one. This would help organizations move forward meaningfully without massive overhauls. Secondly, test the system in a controlled environment before utilizing it in production.

Agentic AI: Regulatory Considerations and Requirements

Auditability and Explainability

Keeping detailed logs by capturing every single agent interaction and system action is needed to maintain auditability. With increasing regulatory requirements, merely showing what the agent did is not enough; the logic and rationale behind each decision must be demonstrable. Quick retrieval of this information is important as banks must be ready for spot checks at any time. These remain key considerations in AI in financial services deployment. 

Risk of Data Residency 

To comply with acts like the GDPR, banks must produce reports detailing how and where client data is stored. Banks that fail to comply with data protection regulations have faced major penalties and bans. In 2021, the RBI banned Mastercard from issuing new cards, as it failed to store payment data locally within India’s borders.

Monitor on a continuous basis 

Static checks are becoming redundant. Regulators are now expecting continuous automated tracking and mediation (if applicable) of output quality and accuracy.

Customer Consent 

Banks are required to explicitly capture customer consent for any data used by Agentic AI. For instance, while the AI has access to your transaction history, it will take customer consent to perform actions such as requesting refunds.

Guidelines for Banks while adopting Agentic AI

While Agentic AI adds several tangible benefits, it is vital that banks follow specific guidelines to ensure they meet regulatory requirements to implement AI in financial services. 

• Make the shift to Agentic AI gradually with a hybrid model comprising both humans and AI. AI can take on a more central role, while humans can focus on maintenance and conducting audits. While humans played a central role in a bank’s operations earlier, Agentic AI will take on that role, shifting the role of humans from executors to supervisors. 

• Restrict the use of agentic AI primarily for complex tasks, after rigorous testing. 

• Take care of compliance proactively. Establishing automated risk assessments and continuous monitoring becomes key to meeting compliance requirements to ensure that AI remains explainable. 

• Robust data governance is critical. This requires clear documentation, tracing every single step and modification made to the data.
  

Conclusion

For the BFSI sector, QA leaders need to take a practical route for agentic AI adoption. Let AI in financial services be rolled out in parts, test it hard, and ensure constant supervision. Trust, explainability, and compliance should be checked every day. While the agents act autonomously, the responsibility for their outputs still lies with human practitioners. The goal of Agentic AI is not to replace people. It was meant to simplify their operations, limiting their work to strategy formulation and implementation. Leading banks have shown that implementing Agentic AI successfully leads to stronger compliance and customer experience. Banks should deploy it with intent and roll it gradually, and the rest will follow.