From AI hype to banking reality

As financial institutions navigate the complex landscape of artificial intelligence, Fabio Strässle, G+D Netcetera's Head of AI Centre of Excellence shares insights on what's driving real value, what's just hype, and how banks can profit from AI transformation. 

Like many people, I was captivated by large language models the moment they emerged – their capabilities and the pace of development have been extraordinary. G+D Netcetera's AI journey goes back nearly a decade, well before the current LLM wave. The team has built immense expertise and technical capabilities over that time. But technical capability alone isn't enough. 

What I hope to bring to the AI Centre of Excellence, drawing on my background as a product manager, is the customer lens – ensuring we're building solutions that solve actual business problems and create real value. Not technology for technology's sake. 

The other realization was: The capabilities of modern AI systems are rarely the bottleneck anymore. The real challenge is making them scalable, secure, and compliant – particularly in regulated industries like financial services. New security challenges are emerging around LLMs and protocols like MCP. Compliance frameworks need to evolve. This is where G+D Netcetera's decades of experience building software to the highest security and compliance standards in regulated industries becomes a genuine advantage. My vision is to bridge that gap: take cutting-edge AI and make it enterprise-ready for financial institutions. 

Traditional AI use cases like fraud detection and risk assessment are proven territory. Where LLMs are genuinely transformative is working with massive amounts of unstructured data: documentation, text, language at scale. Most organizations are drowning in documents. It's often humanly impossible to navigate these knowledge spaces efficiently, and hours are lost searching for specific information. That's typically one of the first areas companies address with LLMs: making sense of unstructured data. Our DocDive solution does exactly this: providing intelligence across documents and data sources, making it easy to find relevant information and get answers to specific questions without hour-long searches.

What's still developing is fully autonomous decision-making. That's where much of the hype – and potential – is concentrated right now. Agentic capabilities are real and growing, but full autonomy isn't there yet. The models have limitations, the human in the loop remains essential, and regulatory frameworks need to catch up.

What's overhyped? The 'plug and play' narrative. The claims that you just plug in a generalized AI tool and everything becomes 50% more productive. Reality is harder: people need skills to use these tools effectively, systems need integration, data needs to be available and high-quality. The traditional expertise of software integration and data governance remains absolutely critical.

Agentic e-commerce is an early trend worth watching. Visa, Mastercard, and others are already exploring agent-to-agent payment protocols and standards. The vision of fully autonomous agents transacting on your behalf is compelling, but there's a long road ahead. Anyone claiming this arrives tomorrow is underestimating the work required.

Honestly, it's too early to claim anyone has the perfect recipe for navigating the AI journey. But we can already identify failure modes that cause real problems. Avoiding those is half the battle.

The first is focusing too much on technology and losing sight of business and customer requirements. Many institutions are jumping on the bandwagon thinking "we have to do something with AI" without first identifying actual problems worth solving. They end up building solutions that seem cool or are engineering-driven based on what's possible with their data, rather than what's actually valuable. That's a trap.

The opposite failure mode is equally damaging: waiting for perfect conditions. Massive “AI ready” infrastructure projects become redundant before they're finished in a space moving this fast.

A smarter approach is partnering with companies where AI is a core business. For financial institutions, AI is a technology to leverage, not their core competency. Building everything from scratch is often a waste of resources. Instead of waiting for perfect conditions, focus on high-value use cases. Build solutions, iterate, get quick wins. Once something works, extend it, scale it, add use cases. Having a vision is good, but the iterative approach generates learnings and maintains momentum – rather than sinking money into projects that go nowhere.

A mindset shift is required: People are accustomed to deterministic software where you get the same output every time. AI is probabilistic. Evaluating AI solutions with the same lens and processes used for traditional software is a failure mode.

Finally, you need the data to be there. AI tools get better with better context. Context engineering is a real differentiator, and that requires a solid data foundation. Without it, even the most capable models underperform.

The honest answer is: you need experts in both, because the two mindsets really are opposite. You can't simultaneously push creative boundaries while being critical and risk-aware. It's like trying to brainstorm and edit at the same time.

At G+D Netcetera, we have people with a long track record in highly secure, privacy-driven industries and our machine learning and AI team pushing the boundaries of what's technically possible. The key is that these groups work together. The innovators and the security experts collaborate rather than operate in silos.

Crucially, security and privacy aren't afterthoughts. They're built into our feasibility stage from the start. When we evaluate whether something is possible with AI, we're simultaneously asking: Can we do this safely? Can we ensure customer privacy? That assessment is integrated from day one.

Risk and real-time fraud protection remains a massive opportunity. Models monitor transactions and generate risk scores to optimize the transaction-to-fraud ratio. You want to cut fraud while reducing false positives – because you want legitimate transactions to go through without adding friction for users. It's a constant balancing act, and it's a huge market. Fraud patterns keep evolving, so there's never-ending potential here.

On the operational efficiency, these new technologies open up many ways to work better with large amounts of information. Tasks that previously required a human – extracting structured data, data entry, analysis, research – can now be significantly automated, freeing employees for higher-value work.

Then there's the customer-facing side. AI banking assistants make it much easier to offer 24/7 support for any question a customer might have. They can provide deep analysis of a customer's data, perform actions, and help users navigate web and mobile banking far more effectively than traditional interfaces.

What's particularly exciting is hyper-personalization on the fly, tailored to individual users in real time. AI unlocks delivering customized product recommendations and experiences to customers at the right moment.

DocDive is an AI platform that orchestrates different AI tools and data sources in an agentic way. At its core, it's a RAG system capable of managing thousands or tens of thousands of documents, allowing users to summarize, search, and extract information easily and efficiently. But it goes beyond that: it can connect to other data sources via MCP in a secure way.

What really differentiates DocDive is deployment flexibility. In highly sensitive industries, Cloud often isn't an option. DocDive can be deployed completely on-premise, with all data remaining on the customer's servers. That provides ultimate control over data, which for many financial institutions is non-negotiable.

The AI Banking Assistant leverages frontier agentic AI capabilities. It doesn't just retrieve information. It can execute actions and transactions through connected tools, securely. This is also a stepping stone toward agentic e-commerce.

What differentiates all our tools, , is deep financial services expertise built into every decision – from architecture to UI/UX. Generic AI tools don't understand banking regulations, data sovereignty requirements, or compliance validation. Ours are built for that reality from the ground up.

It's often said 2026 is the year of agentic AI, and I think we're already seeing this. Over the past year, these capabilities have matured significantly. Standardized protocols are emerging, explorations are happening across industries, and the tools are becoming genuinely useful. We'll see agentic workflows become more stable, predictable, and practical, expanding the range of tasks that can be handled without human intervention.

What's started but isn't yet complete is the development of robust identity and trust frameworks that agents require, especially for payments. Many players are already offering approaches, and we might see standardized solutions emerge by the end of 2026.

We'll see more regulatory clarity and, importantly, industry-driven consensus on safe, privacy-respecting AI use, which often runs ahead of formal regulation.

Looking further is a crystal ball exercise. The speed of innovation is unpredictable. But projecting current trends: agentic AI will become more reliable and autonomous, automating an increasing share of typical tasks. The institutions that will thrive are those building foundations now: investing in data quality, security infrastructure, and the skills to work with these tools. The technology will keep advancing. The question is whether you're positioned to adopt it effectively.

I'm obviously biased, but at this point you really can't close your eyes to this technology. Doing nothing is surely the wrong choice.

I completely understand that the speed and sheer novelty of this is overwhelming. Going at it alone – building up all the expertise internally – isn't the right path for most institutions. My recommendation: find a trusted partner and start small.

What does that mean in practice? Start with concrete business problems, not with the technology. Pick a high-impact use case that provides clear value without massive implementation effort. Quick wins matter: you'll have AI skeptics and inertia, and demonstrating value fast is critical for building momentum.

You need the right data quality, security infrastructure, and compliance framework. Going with a hot consultancy or startup that neglects these fundamentals is risky.  It's easy to focus on building something impressive while neglecting security and compliance – until it becomes a problem.

So to summarize: focus on business-driven solutions. Find a trusted partner and co-design. Choose a high-impact problem where you can reach value quickly. Develop iteratively, learn, and improve continuously.

2015-2018: Foundation years Early AI and machine learning projects across healthcare, mobility, and finance, establishing core expertise in regulated industries.

2019: Industry recognition Named AI partner at World Telematics Innovation Conference. Achieved runner-up position in SBB's Flatland Challenge, solving network disruptions with machine learning.

2020-2021: Healthcare breakthrough Developed Phivea® platform with gMendel for genetic disorder diagnosis using advanced AI and custom deep learning. Awarded third place at Golden Egg start-up competition.

2022-2023: Banking transformation Launched AI banking assistants and DocDive platform. Expanded AI applications across digital banking, fraud detection, and document processing.

2024: Centre of Excellence established Formalized AI Centre of Excellence to centralize expertise, standardize operations, and accelerate business-driven AI solutions across all divisions.

2025-2026: Next-generation AI Focus on agentic workflows, cognitive process automation, and secure AI integration through MCP.