At The Banking Scene Conference 2025 Amsterdam, the message from the panel of industry experts was clear: Artificial Intelligence (AI), and more recently Generative AI (GenAI), is no longer just a buzzword. It's evolving into a set of practical tools with real business value. From regulatory compliance to customer engagement and back-office efficiency, banks are learning how to turn AI from a tech trend into a strategic advantage.

This article brings together insights from a panel featuring ING, de Volksbank, De Nederlandsche Bank (DNB), and nCino, alongside a keynote from Seth Rogers, Director of AI at Kyndryl. Their perspectives go beyond the hype, focusing on practical applications, organisational readiness, and how to drive value with AI, building on the insights we shared in our 2024 white paper on “Generative AI in Benelux Banking” based on interviews with representatives from ING, KBC Bank, De Volksbank, Rabobank, Euroclear and Spuerkeess.

Start with Problems that Matter

"We try to find the biggest problems we can solve, no matter how complex," explained Marco Li Mandri, Head of ING Analytics Office. For ING, these included customer contact centres, KYC, and hyper-personalised marketing. These high-impact areas made it easier to mobilise resources and justify the investment.

"Building a chatbot is easy. Making it safe takes eight months," he said. ING succeeded by focusing on where AI could drive real transformation, and by integrating it into strategic initiatives with clear business goals.

In his keynote, Seth Rogers had also underscored the importance of targeting problems that matter. "Start with two or three areas that really move the needle," he advised. "Let bankers decide where the value lies, not the engineers." He also recommended a stepwise approach: identify big ambitions, but break them down into achievable, incremental wins.

De Volksbank took a similar approach, piloting three internal GenAI use cases. Two are progressing toward production: an internal assistant chatbot and a search tool for regulatory documents. The third, an AI content checker for marketing materials, was paused. "We realised our writing guide wasn’t written for a machine," said Bas van de Werff, Senior Data Scientist. "There was too much tribal knowledge embedded."

The lesson? Not all processes are AI-ready – at least not without a rethink.

Measure What Matters

Across the panel, there was broad agreement that defining success metrics early is key.

"Every AI use case starts with a need and clear KPIs," said Muriël Serrurier Schepper, Programme Manager AI at DNB. At DNB, these KPIs include not just efficiency, but also employee satisfaction and quality of output. "With the war for talent, keeping people happy matters," she added.

ING has adopted a similar model. For example, in KYC, they measure "Straight Through Processing" (STP) rates. In marketing, they track conversion. "You can calculate the 'R' and the 'I', but ROI is tricky because many initiatives overlap," Marco noted.

ING’s solution is to tie AI projects to broader company missions like cost reduction or market growth. If AI helps move the needle, it's valuable – even if you can't assign a precise number.

In response to a question posed from the audience about “lightweighting” Seth pointed out that value can also be measured by operational impact. In one example, he described how a simple shift to on-premise hardware for AI transcription saved a bank €270,000 per month. "Sometimes, optimisation is just about making smarter infrastructure decisions," he said.

Data, Governance, and Getting Organised

As more AI tools proliferate across teams, coordination becomes crucial. In a pre-panel interview, Muriël disclosed that during her first weeks at DNB, she discovered that the organisation was using a wide variety of different AI tools. Part of her current role is to create a centralised approach for evaluating use cases, ensuring ethical oversight, and avoiding duplication.

Seth also advocated for this structured governance model. "We always have humans at the helm," he said. Kyndryl’s framework involves defining foundational ethical principles, implementing central policies, and giving business units flexibility within those guardrails. "We've had great success with this in Germany, where banks operate under varying regulatory environments."

DNB is at the start of that journey. "We don’t have an ethics committee yet, but we’re looking at how others are doing it," Muriël shared. For now, every AI use case must pass three filters: can we do it (from a tech and data perspective), are we allowed to do it (legal and compliance), and should we do it (ethical alignment)?

Seth also warned that AI governance shouldn’t be retrofitted. "Security and identity management must be baked in from the start," he said. He shared examples of banks failing to do this and incurring high costs later.

The Human Factor: AI as a Co-worker, Not a Replacement

Across the board, panellists emphasised the importance of training and culture. ING developed a business training programme for non-technical staff. "Anybody can learn AI," Marco said. "But we needed people who can drive change."

De Volksbank noted that giving employees tools like internal GPT chatbots improved job satisfaction. "People use these tools at home. If they can't use them at work, they might leave," Bas warned, echoing Muriël’s sentiments about the difficulty of retaining staff.

Seth noted the power of AI to simulate human interactions for training purposes. One bank used AI personas to mimic challenging customers, including abusive callers and elderly clients with cognitive impairments. "That reduced call centre training from two weeks to 90 minutes," he said. "It’s not just efficient; it’s also more humane."

For AI adoption to scale, it must enhance employee experience, not threaten it.

Where the ROI is Real

Lara Hörler, Product Success Lead at nCino, sees banks getting the clearest ROI from traditional AI in document extraction and automation. "We typically see 70–80% time savings in spreading financials," she said. That’s an easy win compared to the more unpredictable outcomes of GenAI.

Still, GenAI is quickly catching up. Lara noted its potential to take over once traditional AI has pre-processed the data. Her advice? Start where value is easiest to prove, then layer on more advanced tools.

Seth shared that GenAI is already delivering value in areas like call summarisation and contact centre training. He shared a case where AI-powered simulators reduced agent training time from two weeks to 90 minutes. Another example was integrating AI into KYC processes, allowing a large UK bank to sell onboarding checks to competitors. "The ROI was predicted at six months. It was 30 days," he said.

He also discussed the emerging role of synthetic data and federated learning, both of which enable more privacy-conscious and scalable AI development. "Federated learning lets you train models across banks without sharing sensitive data," Seth explained. "It's already used in Google Photos – now we’re piloting it in financial services."

From Pilot to Production – What It Takes

While many AI projects start strong, few make it to production. ING bucks the trend, with 90% of their AI pilots reaching that stage. Their secret? Executive sponsorship, a central AI platform, and dedicated "AI co-risk" teams that bridge analytics and compliance.

At De Volksbank, pilots are evaluated for efficiency, quality, and staff happiness. And DNB is building a standardised evaluation framework to help teams test ideas responsibly and with clarity.

The lesson: successful AI isn’t about playing with cool tech. It’s about structure, stakeholder alignment, and solving the right problems.

Looking Ahead: What's Next for AI in Banking?

As banks gain experience, they’re getting bolder.

Bas is excited about AI tools that interact with complex data environments. "If AI can connect to the database, read the layers, and show us data lineage, that's game-changing," he said.

Marco sees big potential in hyper-personalised banking, with AI helping determine not just what message to send to customers, but when, how, and via which channel. "It's not just about automation," he said. "It's about resonance."

At DNB, the focus for now is smart scaling. "We're not racing," said Muriël. "We want to be a smart follower, consolidating learnings and rolling out AI in a structured way."

Seth echoed that sentiment, emphasising composable architectures and rapid iteration. "Don’t wait for perfection," he said. "Change is happening every three to six months. Your AI stack should be pluggable and ready to evolve in-flight."

And for Lara? The long-term view is even broader: from fraud detection to end-to-end loan automation and even quantum AI. "It sounds sci-fi, but it's coming," she said.

Final Thoughts

The key takeaway from the panel? AI isn't magic. It's a tool. And like any tool, its success depends on how it's applied, how well it's governed, and whether it's solving a problem worth fixing.

As Marco put it: "You have to find something big enough that if you don't deliver AI, something in the organisation can't move forward."

That’s the mindset that moves AI beyond the buzz.

Join us at The Banking Scene Conference Brussels on May 22 for more practical applications, case studies and insights from industry leaders pushing the boundaries of AI in Banking, in sessions including "Maximise Customer Lifetime Value Through Artificial Intelligence", "Improving Companywide Productivity Through AI Co-Innovation", "From Operations to Customer Care: How GenAI is Redefining Banking", The AI Edge in Payments: Predictive, Personalized, Instant and Secured" and more - don't delay, secure your seat here today!