AI adoption is already mainstream in significant euro area banks
Artificial intelligence is no longer a niche experiment within European banking. In a March 2026 speech on AI and the euro area economy, the European Central Bank reported that nearly 90% of significant euro area banks were already using AI technologies. Banks had also invested more than €4 billion in digital technologies during 2025. Adoption was especially high in fraud and cybercrime detection, followed by marketing, chatbots and credit scoring.
These figures are important because they change the policy and management discussion. The relevant question is no longer whether AI will become significant in banking; it already is. The more pressing questions concern how different applications are governed, what value they create and how institutions manage the risks created by data, models and external providers.
Where banks are using AI
Fraud and cybercrime detection are natural areas of adoption because banking generates large volumes of behavioural and transactional data. Machine-learning systems can identify patterns that traditional rule-based controls may miss, helping institutions prioritise alerts and respond more quickly. Nevertheless, these systems create their own operational issues. False positives may inconvenience customers, false negatives may expose the bank to loss and model behaviour must be monitored as criminal techniques evolve.
Marketing is another common use. AI can analyse customer behaviour and support more personalised communication, but personalisation raises questions about privacy, fairness and the appropriate use of customer data. Banks need to understand not only whether a model improves conversion, but whether the method used is consistent with data-protection, conduct and reputational expectations.
Chatbots and customer-service applications can improve accessibility and reduce routine workload, but the customer may reasonably assume that information provided by a bank is authoritative. Institutions therefore need clear boundaries around the subjects a chatbot may address, escalation routes for complex questions and quality controls for generated responses.
Credit scoring is developing more cautiously because decisions about access to finance are sensitive and regulated. AI can identify patterns that traditional models may not capture, but bias, proxy variables and explainability become significant concerns. Strong validation and human oversight are therefore essential.
Data and model governance become strategic capabilities
The ECB evidence also highlights the importance of data quality. AI projects often expose weaknesses in fragmented information architectures. Models depend on consistent, relevant and well-governed data; otherwise the output may be unreliable regardless of technical sophistication. Investment in AI therefore frequently requires investment in data governance at the same time.
Model governance also needs to evolve. Generative AI behaves differently from many traditional statistical models because output can vary with prompts and context. Testing therefore needs to consider a wider range of behaviour. Banks should understand the circumstances in which a system performs well, the cases in which it fails and the controls that compensate for those weaknesses.
Third-party dependence and operational resilience
Most banks will not develop every foundation model themselves. They will rely on external AI and cloud providers, which creates concentration and supplier risk. Contracts, data-handling terms, service availability and exit options become strategic issues. This is directly connected to DORA and wider supervisory attention to operational resilience.
AI is also entering banks through ordinary productivity products rather than formal model-development programmes. Generative features in office software, coding tools and knowledge systems may be used by thousands of employees. A complete AI inventory should therefore include embedded vendor features, pilots and employee tools rather than only centrally sponsored projects.
Implications for banking skills
As adoption expands, roles will change. Routine analytical and administrative tasks may become increasingly automated, while staff spend more time on exceptions, judgement and customer interaction. This increases the value of professionals who understand both the business domain and the technology influencing their work.
Technology specialists need a strong understanding of banking processes and regulation. Banking professionals need enough AI literacy to challenge outputs and recognise failure modes. Risk and compliance teams need to understand how model behaviour interacts with existing controls. Cross-functional capability will therefore become increasingly important.
The eBSI Banking Academy treats this broader digital competence as part of professional development. Bankers do not need to become AI engineers, but they should understand how increasingly intelligent systems influence risk, customers and operations. The ECB data make clear that this is no longer a future skills question.
Conclusion
AI adoption offers meaningful opportunities to improve efficiency, security and decision support within banks. It also introduces new dependencies, governance requirements and professional skill needs. The institutions that benefit most will be those that avoid both extremes: refusing useful innovation on one side and deploying AI without disciplined controls on the other. Sustainable adoption requires technology, governance and workforce capability to develop together.
Source
European Central Bank, “AI and the euro area economy”, 23 March 2026. The ECB reported that nearly 90% of significant euro area banks already use AI technologies and that realised investment in digital technologies during 2025 exceeded €4 billion.
Adoption figures should be interpreted as a governance challenge as well as an innovation measure
The ECB figures also suggest that supervisors will increasingly judge banks not on whether they use AI, but on whether that use is controlled. When adoption is widespread, AI ceases to be a differentiating novelty and becomes another source of operational and model risk that must be incorporated into existing frameworks. Institutions should therefore ensure that their inventory of AI applications is complete, including functionality embedded in vendor products and ordinary productivity software.
Workforce development should reflect this normalisation. Thousands of employees may interact with AI indirectly without considering themselves AI users. Foundational literacy should therefore reach beyond specialist data teams. At the same time, deeper training should be targeted at employees who validate models, supervise automated decisions or manage customer-impacting processes. A tiered approach is more realistic than assuming one course can meet every requirement across a large bank.
Institutions should also distinguish between experimentation and production use. Pilot activity can be valuable for learning, but once an AI system influences normal banking operations it needs a clear owner, change controls and defined review arrangements. This transition from experiment to production is where informal innovation needs to become formal governance.