- Banks are moving from experimental AI pilots to demanding measurable financial returns, with only 8% having scaled AI beyond initial tests.
- Investment thresholds of 1.5% to 4% of IT spend are critical for meaningful impact, while data quality, skills, and infrastructure remain key barriers.
- European banks see an opportunity in specialized, sovereign AI models, leveraging regulations like the AI Act to build trust and competitive advantage.
- Traditional banks risk losing customer relationships to agile neobanks and tech platforms unless they adopt an AI-first architecture and unified platforms.
The financial industry is undergoing a fundamental shift in how it approaches artificial intelligence. After years of pilot projects and proof-of-concepts, bank executives are now demanding clear, measurable returns on every AI investment. This change is driven by the realization that while adoption is widespread, the ability to convert models into bottom-line results remains concentrated among a handful of institutions. The gap between usage and value is now determining which projects survive the next budget cycle.
According to multiple industry reports, including those from Softtek, McKinsey, and the European Banking Authority, the era of experimental AI budgets is over. Boards are no longer satisfied with vague innovation goals; they want to see a direct link between technology spending and improvements in margins, risk reduction, and revenue growth. This new discipline is reshaping how banks allocate resources and prioritize initiatives.
From Pilots to Profit: The New Mandate for AI Investment
Softtek’s report, The AI-Powered Banking Shift, highlights what it calls the “AI paradox”: technology spreads rapidly, but strategic impact advances much more slowly. Only 8% of organizations have scaled AI beyond pilots or marginal applications with real economic effect. The rest see isolated productivity gains—time savings, better drafting, partial automation—that rarely alter the profit and loss statement of an entire unit. McKinsey’s 2025 global survey reinforces this, identifying about 6% of organizations as “high performers” that attribute at least 5% of EBIT impact to AI. Three-quarters of that group are scaling or have already scaled AI, compared to roughly one-third of others.
The European Banking Authority adds context: 92% of EU banks are already deploying AI, and 55% use generative AI or agentic systems in consumer-facing processes like innovation and invisible risks of fraud detection, customer support, and digital assistants. Yet adoption alone no longer differentiates. The key is whether AI is embedded in core operations or remains a surface-level tool. Softtek estimates that banks need to invest between 1.5% and 4% of their IT budget in AI to cross a threshold where returns become significant. Below that, the return is nearly zero.
This investment threshold is not a universal accounting ratio but a strategic reference. The base of spending, technological starting point, and weight of manual processes vary across banks. Without a homogeneous methodology, two institutions could report similar percentages while funding very different capabilities. The potential value, however, is enormous: Softtek projects between $200 billion and $340 billion annually for global banking, equivalent to 9%–15% of operating profits. But that future value requires linking each initiative to concrete levers—cost per transaction, conversion rates, fraud losses, resolution times, credit accuracy, revenue per customer, or capital consumed.
Data, Skills, and Infrastructure: The Building Blocks for Scaling
One of the biggest obstacles to scaling AI is incomplete or low-quality data. Ernst & Young’s Adam Smith notes that AI needs quality data but can also help improve it. His firm assisted a large bank in using AI to understand and verify data used in credit underwriting, boosting accuracy to about 90% and allowing employees to focus on likely error areas. Banks are exploring AI-based tools for data validation and compliance, though many are still in development. A unified platform approach, as advocated by EY’s Cox, prevents siloed AI agents that limit process integration and customer service improvements.
Skills are another critical barrier. When asked about challenges in generating value from autonomous AI, 58% of banks cited a lack of technical knowledge and capabilities. Sameer Gupta of EY notes that banks need three to five times more staff than five years ago, including AI engineers, data scientists, application developers, and cybersecurity specialists. Training the entire workforce to use AI tools confidently is essential, combining specific courses with change management programs.
Infrastructure decisions are also evolving. As AI workloads grow, computing power demands skyrocket. Some banks prefer cloud scalability; others adopt a hybrid approach. Niranjan Vivekanandan of RBC Commercial Banking explains that building their own GPU infrastructure provides greater security, privacy, and sovereignty—critical for trust. However, the high upfront cost makes this viable only for the largest banks. Meanwhile, cloud pricing models are shifting, and vendor dependency raises strategic concerns.
The Competitive Landscape: Who Will Own the Customer?
AI is systematically eroding traditional barriers to entry in banking. Softtek warns that tech platforms, neobanks, and smart finance providers can capture the customer interface and recommendation layer, leaving traditional banks as mere balance sheet providers. These new entrants operate without legacy systems, using cloud services and contracted components to target profitable segments. Incumbents retain hard-to-replicate assets—transaction history, scale, trust—but must convert them into faster decisions and coherent experiences.
Europe has a unique opportunity to lead in specialized, sovereign AI. The Santander AI Lab and CSIC report, La nueva frontera europea de la IA B2B, based on analysis of 10.3 million documents, predicts that the next generation of B2B AI will be marked by specialized models, data sovereignty, and regulatory trust. Small Language Models (SLMs) and sovereign RAG architectures allow organizations to keep corporate knowledge under control while leveraging generative AI. Frameworks like the AI Act, DORA, GDPR, and NIS2 become competitive advantages for those that integrate governance and risk management from the design stage.
José Manuel de la Chica, director of Santander AI Lab, emphasizes that Europe cannot compete with the US or China on foundational model investment—the US mobilized $109 billion in private AI investment in 2024 versus Europe’s $19.42 billion—but can lead in specialized AI for regulated sectors. “The advantage will not be in always using the most powerful model, but in integrating the right AI into the right process, with traceability, data control, and responsibility from design.”
The competitive window is finite. Israel Quiñonero Fernández, Softtek’s Director of Technology for Banking in EMEA, warns that the opportunity to build sustainable AI advantages is real but not infinite. Late investment can buy infrastructure but takes longer to rebuild processes, talent, and operational discipline. The 2027 budget discussions will treat AI as an investment portfolio, not an innovation program. Each use case must have a business owner, baseline, total cost, acceptable risk, and review date. Some pilots will close; others will receive more capital precisely because they require deep changes in data and legacy systems. The challenge for banks is to finance this transformation without turning urgency into indiscriminate spending or stalling projects whose returns only appear at scale.
Ultimately, the shift from experimentation to execution is reshaping the entire banking industry. Banks that successfully integrate AI into their core operations—with unified platforms, quality data, skilled teams, and robust governance—will be the ones that capture the $200–340 billion annual value at stake. Those that fail to move beyond isolated pilots risk being left behind as nimble competitors redefine customer relationships and cost structures. The next few years will separate the leaders from the laggards in the AI-powered banking revolution.
