About Andrei Serjantov: Head of Digital Global Markets at BNP Paribas CIB, leading digital transformation and AI integration within the investment banking sector.
About BNP Paribas CIB: The corporate and institutional banking arm of BNP Paribas (approx. €46B Group Revenue), serving 13,000+ corporate and institutional clients worldwide.
About Nathalie Beslay: CEO and Co-Founder of naaia, a RegTech entrepreneur specializing in helping organizations manage AI projects within strict regulatory constraints.
About naaia: A specialized technology company providing a SaaS platform (AIMS – AI Management System) for AI governance, compliance, and oversight; focuses on bridging the gap between static regulation and dynamic AI agents.
About Cyril Cymbler: Head of Financial Services EMA at Databricks, expert in transforming raw financial data into “Data Intelligence” to drive value.
About Databricks: A global leader in data and AI (Data Lakehouse), valued at ~$43B (2023), aiming to unify data, analytics, and AI for enterprise use cases.
Why is there an urgent need to reinvent trust and data strategy in finance now?
Cyril Cymbler: The necessity is driven by three factors. First, the exponential growth of structured and unstructured data requires managing massive workloads to build strong agents. Second, the industry is shifting from “General Intelligence” (standard ChatGPT, ~55% accuracy) to “Data Intelligence” (structured analysis engines, >80% accuracy/conversion). Third, governance is non-negotiable; without strict data structure, even the best ML models fail due to “Garbage In, Garbage Out.”
Andrei Serjantov: The drive is about efficiency, scope, and capability. AI allows banks to execute existing tasks faster, cover a broader spectrum of risks (covering risks A, B, and C instead of just A), and perform previously impossible tasks, such as analyzing massive document volumes for sentiment analysis in equity research.
What are the concrete, high-impact use cases being deployed in Investment Banking?
Andrei Serjantov: Beyond generic utilities (coding, translation), BNPP focuses on deep vertical integration. In Equity and Fixed Income research, there are over a dozen use cases transforming the document production pipeline. Client workflows are heavily augmented, where LLMs draft answers to client trade confirmations. The strategy is pervasive: it is now harder to find areas not using AI than those that are.
How do organizations handle the specific challenges of “Agentic” AI and static Regulation?
Nathalie Beslay: The core conflict is that regulation (like the EU AI Act) is static, while Agentic AI is dynamic. The challenge lies in qualifying agents against strict regulatory definitions (General Purpose AI vs. AI Systems). Furthermore, because agents are autonomous, companies must rigidly manage roles and “habilitations” (permissions) to ensure agents interact safely with human and digital environments.
How should a bank technically execute a “Zero to One” AI platform reinvention?
Cyril Cymbler: Execution requires a three-pillar business approach: Growth, Protection (Risk/Compliance), and Cost Efficiency. Technologically, the platform must be Multi-cloud (to satisfy DORA regulations on data portability), Open Source (ensuring data ownership/zero-copy), and Governed (strict data lineage to track product creation for auditors). Finally, AI must be democratized, allowing business leaders to query data in natural language without relying on technical specialists.
How do you operationalize “Trustworthy AI” (Compliance, Ethics, Responsibility)?
Nathalie Beslay: Trustworthiness is built on three layers: Compliance (Mandatory), Ethics (Human-centric), and Responsibility (Accountability). naaia operationalizes this via a SaaS approach focusing on Governance (designing roles/decision trees), Integration (embedding testing/monitoring into IT systems), and Execution (enforcing documentation). Passive complaint is not a strategy; proactive program deployment is required to mitigate economic and reputational damage.
What is the strategy regarding “Build vs. Buy” and partnerships (e.g., Mistral)?
Andrei Serjantov: BNPP partnered early with Mistral to leverage On-Premise capability, ensuring customer data never leaves the bank’s secure environment. They prioritize fine-tuning smaller, energy-efficient models on proprietary data over training massive foundational models from scratch. This combination of external tech expertise and internal risk management DNA allows for secure, specific use case deployment.
