At TBSCONF25BXL on May 22, an all-star panel of industry leaders unpacked how generative AI is reshaping the entire banking value chain: from operational efficiency and internal productivity to customer engagement and strategic governance.

The message was loud and clear: GenAI is no longer a playground for pilots, it's a catalyst for real transformation.

I had the pleasure of moderating a panel featuring:

• Marcia de Wachter, Chair of the Board, MeDirectBank Belgium
• Anthony Belpaire, Head of AI, BNP Paribas Fortis
• Christophe Atten, Team Manager Data Science & AI, Spuerkeess
• Sophie Dionnet, SVP, Product and Business Solutions, Dataiku
• Swaraj Poonye, Senior Capability Lead, Expleo

Together, they provided a practical glimpse into how GenAI is being operationalised at scale and in production across Europe’s banking sector.

Building AI-Ready Organisations Starts at the Top

One of the standout themes of the discussion was the importance of leadership commitment and culture when embedding GenAI capabilities into the business. Marcia de Wachter of MeDirectBank was unequivocal:

“You have to walk the talk... If you don’t know what AI is, how can you judge its value? Then you are sailing blindly.”

At MeDirect, this commitment to understanding and using AI started with every single employee, from board members to facilities staff, undergoing GenAI training encompassing basic understanding to building their own GenAI models! This comprehensive, hands-on training was cited as a key factor in driving internal acceptance and AI fluency across the bank.

Anthony Belpaire of BNP Paribas Fortis echoed this view, revealing that over 11,000 employees at BNP now use an internal AI assistant daily:

“We probably are one of the most advanced labour forces in terms of AI literacy, and you see lots of bottom-up innovation, because who knows better where you could potentially optimise the process than the people doing the work themselves?”

AI adoption is even tied to the bonus and remuneration structure at BNP, reinforcing its strategic significance.

Operating Like a Start-Up at Scale

One of the most unique models presented came from BNP Paribas Fortis, where the AI team is structured as an internal scale-up with its own AI board and delivery autonomy:

“We run like a scale-up inside the bank,” said Anthony. “We define roadmaps, build and run products, and measure value in terms of annual recurring value, not revenue, but tangible business impact.”

This setup has enabled fast rollouts, cross-functional alignment, and a ruthless focus on measurable outcomes across three strategic pillars: AI for clients, AI for processes, and AI to augment employees.

The Open Source Advantage

While many banks still rely on major vendors, Spuerkeess has charted a different path. Christophe Atten shared how Luxembourg’s state-owned bank has gone all-in on open-source LLMs hosted entirely on-prem:

“We don’t need cloud for this. Small, open-source models can do exactly what we need, for HR bots, user stories, even internal translation, and we can scale them safely.”

This approach has enabled them to control cost, increase trust, and maintain compliance with national data regulations, while still delivering meaningful productivity gains.

Spuerkeess measures GenAI success in one of its early projects by the reduction in queries sent to HR, usage rates across departments, and feedback loops to improve coverage and accuracy.

Structuring for Scale: Decentralised Innovation, Centralised Trust

Sophie Dionnet from Dataiku highlighted the shift in how AI innovation is structured from a people perspective:

“GenAI forces organisations to rethink what is deemed an acceptable use of technology. It's forcing an opening of putting stuff in the hands of people, even if they are not initially considered as being the right type of people to be equipped with a data tool.”

However, she warned against going too far without structure:

“We see customers being successful when governance is not an afterthought. It's the cornerstone of trust and the enabler for scale.”

This sentiment was reinforced during a broader discussion on data quality. GenAI is amplifying awareness of weak data foundations, forcing companies to clean up their digital backyards:

“Before, if something went wrong, you could blame a human. Now, people start questioning their data and that’s a good thing,” Sophie noted.

Ethics, Regulation and Responsible AI

GenAI may be powerful, but it also introduces new risks. Ethical considerations, particularly in regulated sectors, were front and centre in the panel.

Marcia de Wachter detailed how MeDirect incorporates compliance and fairness into its AI models through policy translation, model testing, and internal auditing:

“Everything from MiFID to internal ethical policies is encoded into the AI systems. We test answers for compliance and bias, and continuously refine them.”

Swaraj Poonye from Expleo added that the real work in ethical AI is in operationalising values:

“Ethical AI is more than good intentions. We help banks break down fairness, trust and bias into questions they can answer and test.”

Both panellists acknowledged the complexity of aligning legacy processes, vague ethical policies, and new algorithmic systems. Still, the EU AI Act was seen as a net positive, a force to instil better governance.

Agents, RPA on Steroids, and the Road Ahead

A notable shift is underway from traditional RPA to more intelligent, autonomous, agentic systems.

“Agentic is just the new buzzword,” said Christophe Atten. “But really, it’s just smarter workflows, with unstructured data, chaining, and decision-making in the loop.”

Christophe envisions a future where employees are supported by multiple virtual agents, each handling a slice of their repetitive tasks, freeing up humans to think strategically and creatively.

Sophie Dionnet was cautiously optimistic:

“Agents can be a catalyst for everything else, your data, your models, your business rules. But their value depends entirely on the quality of your foundations.”

Build, Buy or Partner? And What About Quantum?

Anthony Belpaire touched on the build vs. buy dilemma:

“We don’t build LLMs ourselves; we partner. But we’re selective, because partnering with startups can be painful in banking. Where speed or capability is critical, we do it.”

BNP Paribas has signed strategic partnership agreements with firms like Mistral to stay close to frontier development, while building domain-specific agents in-house on trusted infrastructure.

And what of the next frontier: quantum?

“Quantum today is like AI was in 2010,” said Swaraj. “Promising, but still evolving. Banks should keep an eye on it, but right now, focus on GenAI.”

Final Thought: GenAI Is the Train You Have to Catch

Marcia de Wachter provided a powerful metaphor:

“GenAI is the train you have to catch. It’s more important than even the World Wide Web.”

All panellists shared that sense of urgency. Whether you're a top-tier incumbent or an agile challenger bank, generative AI is redefining customer expectations, reshaping internal operations, and rebalancing what technology strategy looks like in banking.

But catching that train isn't just about experimenting with tools. It requires:

• Executive-level commitment
• Thoughtful governance
• Robust data foundations
• A culture of learning
• A willingness to partner and adapt

GenAI is not a silver bullet, but it is a strategic lever. Banks that invest in making it work today will define the competitive landscape tomorrow.