Banks Want More AI: Are They Ready for What Comes Next?

Artificial intelligence is entering fraud detection, anti-money-laundering systems, customer service, cybersecurity, document processing and internal decision-making. However, the harder question is shifting from whether banks can deploy AI to whether they can govern it once it becomes embedded in critical financial processes.

New study "Artificial Intelligence Governance and Organizational Readiness in Banking: Evidence from Albania," published in the journal Administrative Sciences, by Laureta Domi, Majlinda Godolja and Kozeta Sevrani of the University of Tirana, examines how technological readiness, institutional governance, perceived benefits and implementation barriers shape AI adoption and scaling across Albania's banking system.

Based on responses from 85 professionals across 15 institutions, including all 12 banks operating in Albania and three additional financial institutions, the research provides a useful window into a challenge that extends far beyond one small European market: AI adoption can move much faster than the organizational systems required to control it.

AI Has Entered the Bank, but It Has Not Yet Rebuilt the Bank

Some 27.1% of respondents reported active AI use in production, while another 28.2% reported pilot or limited deployment. Combined, 55.3% indicated active or pilot AI use, while a further 28.2% said adoption was planned. However, deployment is not the same as institutional transformation.

Only 4.7% of respondents described AI as broadly integrated, while just 3.5% reported full or end-to-end integration. Regular monitoring was reported by 17.6%. The study also finds that governance structures and benefit measurement remain uneven. These numbers expose the gap between having AI applications and possessing mature systems for managing them.

An experimental productivity tool can fail quietly. An AI system affecting creditworthiness, transaction monitoring, fraud alerts or compliance decisions can generate consumer, regulatory and financial-stability consequences. The study's AI Maturity Index reinforces this picture. The average score was 49.0 out of 100, but results varied widely. Data and Technology Readiness averaged 60.7, while adoption and integration scored only 34.8. In other words, some of the technological foundations appear to be developing faster than deep operational use.

Forty percent of respondents were classified in the low-maturity category, 38.8% in moderate maturity and 21.2% in high maturity. More revealingly, 31.8% demonstrated scaling ambition above their current maturity profile. The danger is clear: institutions may be preparing to accelerate AI before their governance, integration and monitoring systems are fully ready.

The Real AI Bottleneck Is Becoming Organizational, Not Technological

Generative AI was the most commonly reported use case, cited by 35.3% of all respondents and 63.8% of those reporting active or pilot AI use. Process automation, customer-service tools, fraud and transaction monitoring, cybersecurity and document processing were also prominent. Yet when respondents were asked about barriers, the biggest constraint was not access to technology. It was people and institutional capacity.

The skills or capability gap was reported by 57.6% of respondents, followed by regulatory or compliance uncertainty at 45.9%, cost at 44.7%, integration with existing systems at 43.5%, and ethical, transparency or reputational concerns at 40%. Data-quality and third-party risks were also significant.

This changes the strategic conversation around banking AI. The next competitive divide may not be between institutions that have AI and those that do not. It may be between institutions capable of embedding AI within robust organizational systems and those accumulating disconnected applications without sufficient governance.

The study's statistical results strengthen that interpretation. Data and Technology Readiness was positively associated with Capability–Governance Readiness. Environmental and regulatory pressure also showed a positive relationship with governance readiness in the main model. In turn, stronger Capability–Governance Readiness was associated with higher Perceived Benefits, while Perceived Benefits were strongly associated with intentions to invest further or scale AI.

Governance is not merely a defensive compliance function. It may form part of the institutional capability through which banks identify valuable AI applications, coordinate deployment and gain confidence to expand successful systems.

Regulation Could Become an Enabler of AI Maturity, Not Just a Constraint

Technology debates often present regulation and innovation as opposing forces. The study suggests a more complicated relationship. In banking, external regulatory and supervisory pressure may encourage institutions to clarify responsibilities, improve documentation, strengthen monitoring and build risk-management capacity. The Albanian case is particularly instructive because its banks operate in an environment increasingly influenced by European and international expectations around cybersecurity, outsourcing, data protection, operational resilience and accountability.

For supervisors, the practical lesson is not necessarily to produce more abstract AI principles. Banks need usable governance infrastructure.

The researchers highlight measures such as AI-use-case inventories, clear ownership of systems, documentation requirements, model and vendor oversight, human-review rules, monitoring routines and incident-reporting expectations. They also argue for cross-functional governance involving technology, business, risk, compliance, audit and senior management rather than leaving AI exclusively to IT teams.

The recommendation is especially relevant for smaller banking systems and emerging markets. Large global banks can devote substantial resources to model-risk teams, AI governance committees and technical specialists. Smaller institutions may rely more heavily on external vendors and cloud-based systems while having less capacity to independently validate what those technologies are doing, creating a development-policy challenge

Wider access to AI can reduce technological inequality, but access without institutional capability can create a new governance gap. Investment in digital transformation must therefore include supervisory expertise, workforce training, data governance and audit capacity, not just software procurement.

The Next AI Race: Proving Value Without Losing Control

The researchers expected stronger perceived barriers to reduce intentions to scale AI. The data did not support that relationship. The coefficient was positive rather than negative and statistically insignificant. The authors appropriately caution against interpreting this as evidence that obstacles encourage AI expansion. A more plausible explanation is that institutions already engaged with AI are more aware of its practical difficulties.

The insight captures the current stage of AI adoption particularly well. Banks can simultaneously believe AI is strategically important and recognize that they face serious problems involving skills, legacy systems, regulatory uncertainty, data quality and vendor dependence.

The more important predictor of scaling intention was perceived benefit. It puts benefit measurement at the center of responsible AI strategy. Banks need evidence that applications reduce processing time, improve fraud detection, strengthen customer service, enhance compliance or generate other measurable outcomes before moving from pilots into core systems. Otherwise, AI expansion risks becoming technology accumulation rather than productive transformation.

The study remains exploratory. Its 85-person sample is modest, its measures are self-reported and its cross-sectional design cannot establish causality. The responses represent informed professional perceptions rather than independently audited institutional assessments. The authors also acknowledge that common-method bias cannot be entirely ruled out. Future research will need longitudinal evidence, cross-country comparisons and stronger links between survey perceptions and objective deployment, audit and performance data.

However, those limitations do not weaken its key policy message. For Albania, and potentially for many smaller and developing financial systems, the crucial phase of AI transformation is arriving after the first wave of adoption. Banks now have to move from experimentation to institutionalization. It means building organizations capable of knowing where AI is used, who is accountable for it, how its performance is monitored, how customers are protected, how vendors are controlled and whether the technology is actually creating value.

The winners in banking's AI transition may therefore not be the institutions that deploy artificial intelligence fastest, but may be those that learn how to scale it without allowing technological ambition to outrun institutional control.