Table of Contents
Key Insights from Industry Leaders
The financial services industry stands at a transformative crossroads where artificial intelligence is evolving from simple automation to sophisticated, autonomous decision-making systems. On July 9, 2025, leading CTOs, CIOs, and fintech innovators from across the Middle East and Europe gathered virtually to explore the future of Agentic AI in finance. The discussion revealed both the immense potential and the significant challenges that lie ahead in implementing truly autonomous financial systems.
Defining Agentic AI: Beyond Simple Automation
Mikhail Khasin, Head of Global Markets Technology at BNP Paribas Portugal, provided a foundational understanding of how organizations should approach Agentic AI implementation. Drawing from his experience at Europe’s large banking institutions, Mikhail emphasized that Agentic AI represents an evolution in organizational maturity with AI systems.
“Think of LLMs as bachelor graduates from different universities,” Mr. Khasin explained. “Each vendor trains their model based on their own curriculum, but none of them knows your organizational context.”
He outlined a structured approach to Gen AI maturity:
- Foundation Layer: Start with large language models (LLMs) from various vendors
- Contextualization: Build RAG (Retrieval-Augmented Generation) databases with organizational knowledge
- Orchestration: Move from single prompts to sequences of AI agents working together
Mr. Khasin illustrated this with a practical example from software development:
“You can create agents that generate code from requirements, test that code, analyze bugs, and iterate through multiple rounds of fixes until you get clean, functional code. This is how you avoid the hallucinations and reliability issues that plague single-prompt implementations.”
The African Perspective: Trust and Transformation
As African financial institutions continue their digital evolution, the adoption of advanced technologies like Agentic AI presents both transformative opportunities and complex challenges—especially within the context of customer trust and regulatory compliance.
Pragashani Reddy, Executive Director of Digital at Business Banking, Absa Group, brought insights from one of Africa’s largest financial institutions. Her perspective highlighted the unique challenges and opportunities in implementing Agentic AI across diverse regulatory environments.
Agentic AI, which refers to artificial intelligence systems capable of making autonomous decisions or taking initiative within defined boundaries, is reshaping the customer experience across the banking sector. Speaking at the Fintech AI conference on July 9, Pragashani Reddy shared her insights into how Agentic AI can drive hyper-personalization in financial services. “Post-pandemic, anticipating customer needs has become critical,” Reddy noted. ‘An AI agent might detect signs of financial distress and proactively offer restructuring options, moving us from transactional to relational banking.’
Yet, the promise of Agentic AI cannot be fully realized without addressing the crucial dimension of human oversight—especially in high-risk scenarios. Reddy emphasized the need for responsible AI implementation, stating: “For trading and large funds movements, the risk factor is too high for fully autonomous agents. We focus on advisory roles where agents guide decisions, but the final transaction remains with the customer.”
The keynote further explored how trust is central to the future of Agentic AI in banking. For customers to accept and engage with AI systems, they must trust the logic and ethics behind each decision. “Trust hinges on transparency, explainability, and control,” Reddy said. “Customers need to understand the rationale behind agent actions, particularly in sensitive areas like credit decisions and investment advice.”
In an African context—where digital inclusion, data privacy, and legacy financial inequalities intersect—the path to responsible Agentic AI must be paved with cultural sensitivity, robust governance frameworks, and clear accountability. Reddy’s insights serve as a call to action: to harness the power of AI without compromising human agency, dignity, or financial well-being.
As institutions across the continent advance toward more intelligent systems, building trust through transparency and ethical design will remain at the heart of AI-powered transformation.
European Regulatory Perspective: Sovereignty and Security
Nena Dokuzov, responsible for Digital Transformation of the Economy at Slovenia’s Ministry of Economy, Tourism and Sport, provided a cautionary yet forward-looking government perspective on Agentic AI implementation.
“I am very much reserved towards generative AI in autonomous finance,” Dokuzov stated frankly. “It’s too risky from a data protection and disclosure standpoint.” However, she outlined a path forward through what she termed “AI continuum” – a European approach to distributed AI that addresses sovereignty concerns.
Her vision centers on several key principles:
- Lower latency through edge computing bringing AI closer to data sources
- Reduced costs for data access and processing
- Enhanced security through encryption and zero-knowledge protocols
- Sovereign control over AI models and data processing
“We need to find the right balance between blockchain’s zero-knowledge approach – where the middleman doesn’t know the information – and agents that need to remember things,” Dokuzov explained. This represents a fundamental tension in designing secure, autonomous financial systems.
Moderator's Perspective: Bridging Strategy and Execution
By Ankur Handoo, Co-founder of Hueman AI & Event Moderator
“Guiding this incredible panel discussion revealed a fundamental, productive tension at the heart of autonomous finance. On one hand, we heard from banking leaders like Mikhail Khasin and Pragashani Reddy about the immense strategic importance of organizational context and building customer trust through transparency and human oversight. On the other, technologists and builders showcased powerful, ready-to-deploy use cases in areas like AML and compliance.
The central question that emerged was not if we should adopt Agentic AI, but how we bridge the gap between high-level strategic imperatives and the practical, on-the-ground reality of implementation. As Nena Dokuzov rightly pointed out, issues of data sovereignty and security are paramount, while the 80% failure rate for AI pilots, mentioned by Ahmed Osama, looms large.
Implementation Challenges: The 80% Failure Rate
The discussion took a sobering turn when addressing the widely cited prediction that 80% of enterprise AI pilot projects will fail. This statistic, rather than signaling a failure of AI itself, underscores deep-rooted issues in implementation strategies.
Arun Pandit, CTO of Hueman AI, emphasized that most failures stem not from technology limitations, but from flawed execution. “To move from a promising pilot to a scalable, production-grade system,” he stated, “organizations must adopt an architecture-first, inside-out approach.”
Drawing from both his experience and the broader panel’s insights, Arun outlined three key principles:
- Architecture Over Models: The focus should shift from selecting the “best” LLM to designing the right architecture around it. Echoing BNP Paribas’ Mikhail Khasin, Arun noted that LLMs lack organizational context by default. At Hueman AI, success comes from building robust systems for orchestration and context engineering—such as dynamic RAG databases and auditable agent behavior—ensuring compliance, transparency, and trust, a point also raised by Pragashani Reddy.
- The Inside-Out Mandate: Organizations cannot expect AI to fix processes they don’t fully understand. The prerequisite is mapping and structuring workflows for AI consumption through semantic vector mapping. This foundational step prevents hallucinations and turns a generic LLM into a reliable, domain-specific expert.
- Iterative Autonomy with Human-in-the-Loop: Full autonomy, especially in high-risk areas like trading or compliance, is not the immediate objective. Instead, phased implementation with human-in-the-loop architectures builds trust and ensures responsible adoption. Agents begin in advisory roles, gradually assuming more responsibility as they prove reliable—an approach advocated by both Aditya Gupta and Ahmed Osama.
Ahmed Osama, Head of Artificial ntelligence of Banque Misr added a practical lens to the discussion, urging firms to adopt agile methodologies and hybrid architectures. “The key is starting with a lean approach,” he said. “Avoid regulatory complexities initially so you can measure real value with the smallest implementation possible. Use agents to build the MVP of other agents – iterate fast, get feedback, and kill projects quickly if they don’t deliver ROI.”
He also reframed how organizations should evaluate return on investment (ROI): “It’s not just financial. Improvements in customer experience, operational efficiency, or sustainability can all be valid ROI metrics.”
Complementing this, Aditya Gupta, CEO at Dotnitron Technologies addressed data security head-on, highlighting two main approaches:
- On-premises deployment of open-source models—expensive but secure
- Data abstraction layers that anonymize user-specific data before sending it to external LLMs
Together, these perspectives reveal that successful AI implementation is as much about organizational design and engineering as it is about data science. By focusing on scalable architecture, process transparency, lean experimentation, and pragmatic risk management, enterprises can beat the 80% odds and build AI systems that deliver real, sustained value.
Practical Use Cases: Where Agentic AI Shines Today
Several speakers highlighted specific areas where Agentic AI is already delivering value:
- Regulatory Compliance and AML
Aditya Gupta described a compelling use case in Anti-Money Laundering (AML): “Agents can pull policy text, check it against case data, map it back, and provide citations explaining their decisions. This transforms manual compliance work into an explainable, auditable process.” - Customer Onboarding
Arun Pandit from Hueman AI noted successful implementations in digital customer onboarding: “Major banks like ICICI in India have automated customer onboarding using Agentic AI systems that integrate customer verification platforms with computer vision for document processing.” - Portfolio Management
Mohammed Algohary from Packtech outlined applications in loan origination and portfolio management: “Agents can autonomously maintain portfolios aligned with organizational objectives, continuously monitor compliance with regulatory requirements, and refine criteria based on real-time market conditions.”
The Road Ahead: Organizational Design and Human-AI Collaboration
Arun Pandit emphasized that successful Agentic AI implementation requires fundamental organizational redesign: “We cannot expect customers to fill gaps that haven’t been identified by organizations themselves. The responsibility lies with system designers and architects to look inside our own organizations first.”
His framework for AI-native organizations includes:
- Context Engineering: Ensuring organizational memory and processes are properly documented and accessible to AI systems
- Human-in-the-Loop Architecture: Starting with human oversight and gradually increasing autonomy
- Audit Trails: Maintaining complete records of AI decision-making processes
- Semantic Vector Mapping: Properly structuring organizational knowledge for AI consumption
Key Takeaways for Fintech Leaders
The panel surfaced five critical takeaways for decision-makers implementing Agentic AI:
- Start Small, Think Big
Begin with controlled, low-risk scenarios such as compliance checks or support automation. Once maturity builds, scale to more complex functions like trading or underwriting.
- Context Is Your Advantage
Owning the latest AI model isn’t enough. What matters more is how well the model understands your organization’s context—processes, data, and domain-specific nuances.
- Explainability Is Non-Negotiable
Whether for compliance, transparency, or customer trust, every decision made by an AI agent should be traceable and understandable.
- Plan for Sovereignty
Make data ownership, model control, and regulatory alignment foundational to your implementation strategy. This is especially critical in cross-border operations.
- Broaden ROI Metrics
Financial return is only one aspect. Measure success by how much your AI improves customer satisfaction, reduces risk, or streamlines operations.
Conclusion: The Future is Collaborative, Not Fully Autonomous
The most powerful insight of the event was that Agentic AI is not meant to replace human expertise but to elevate it. The future of finance lies in creating systems where AI agents handle complex tasks while humans retain strategic oversight and decision rights.
Rather than striving for complete autonomy, the goal is to build AI ecosystems that enhance responsiveness, personalization, and resilience across the customer journey. In this vision, AI agents become intelligent collaborators—augmenting teams, reducing friction, and unlocking new possibilities for financial empowerment.
This shift will require organizations to rethink everything from governance to system design. But those who get it right won’t just be ahead in technology—they’ll redefine how financial value is created in the digital age.
Post Tags:
Share: