Here, you’ll find comprehensive guidance on transforming your data for better analysis, ensuring it’s clean, standardised, and insightful. Begin your journey towards expert data handling now.
Key takeaways
- AI adoption in banking has moved past the pilot stage, but scaling remains the unsolved problem, with only a fraction of institutions achieving enterprise-wide deployment.
- Resilience and control are replacing productivity as the primary drivers of AI infrastructure decisions, shaped by DORA, the EU AI Act, and growing data sovereignty concerns across European markets.
- The highest-value AI applications in financial services organisations (fraud detection, credit decisioning, document processing) are already proven; the gap is in the infrastructure and governance needed to run them reliably at scale.
- The institutions pulling ahead are solving the foundational problems: data integration, AI governance built in from the start, and infrastructure they own and control.
The state of AI adoption in banking: past the pilot stage, short of scale
The numbers suggest significant progress. AI is already used by 33% of UK financial services companies for fraud detection. According to an EY-Parthenon survey of 100 senior decision-makers across retail and commercial banking, 47% of respondents had fully implemented GenAI applications by early 2025, compared to just 10% in 2023. That’s a meaningful shift reflecting genuine investment across the sector.
But implementation and scale are not the same thing. Research from MIT found that only 5% of integrated AI pilots have delivered significant value and been embedded at scale into live workflows. The gap between those two figures is where most banks currently sit: invested, but not yet transformed.
The barriers are structural rather than strategic. Core legacy systems, fragmented, unstructured data, and compliance constraints slow deployment at every stage. Nearly 30% of AI pilot projects in banking never reach production, most often due to unclear ROI or unresolved regulatory questions. What is changing in 2025 is where institutions direct their focus: rather than broad automation goals, banks are applying AI and automating processes in specific, high-friction workflows, particularly in lending, onboarding, and document-heavy financial operations to deliver real-time insights to finance teams. What AI also improves is strategic decision-making through accurate forecasting and analysis.
The ambition has become more targeted, and with that, more executable. However, targeted deployment is still a long way from enterprise-wide adoption that would constitute a genuine structural advantage. The institutions that close that gap will not do so by running more pilots, but by solving the infrastructure problem underneath. And increasingly, that means asking not just what AI can deliver, but whether the architecture sustaining it is resilient enough to scale on their own terms.
Enhancing the architecture of an application that enables over 20 million invoices to be processed each day
Read the case studyWhat AI is actually being used for in financial services
The conversation about AI adoption in finance has matured past the theoretical and triggered a fundamental shift. ECB Banking Supervision data points to a strong increase in AI deployment among European banks between 2023 and 2024, with credit scoring and fraud detection leading adoption.
Beyond those two, document processing, customer onboarding, portfolio management, and regulatory reporting are all seeing meaningful integration, each driven by different operational pressures, each producing different types of measurable return. What follows is a look at where those returns are already visible.
Automating the document-heavy middle office
The operational backbone of a financial institution is built on documents: vast amounts of financial data, contracts, onboarding forms, KYC files, credit applications. Most of it has been processed manually for decades, which means it is slow, error-prone, and difficult to scale.
According to Fenergo’s 2025 Financial Crime Industry Trends Report, AI adoption in KYC and AML processes surged from 42% in 2024 to 82% in 2025, with average compliance spend per firm reaching $72.9 million, a figure that gives a clear sense of the cost pressure driving automation decisions.
In the financial sector, invoice intake and factoring operations sit at the same intersection of volume and complexity. When document flows are high and data quality is inconsistent, which is unfortunately the norm rather than the exception in supply chain finance, the cost of manual handling increases quickly.
AI-driven intake systems address this by creating a single, structured entry point for document data: extracting, validating, and routing information without human intervention for standard cases.
For our clients in factoring and invoice discounting, this has translated into up to 50% faster invoice processing, 40% straight-through processing for standard cases, and a reduction in manual effort of up to 30%.
When document processing becomes reliable and auditable, it changes what downstream decisions are possible and how quickly they can be made.
Improve your financial operations with AI and automation
From invoice intake to decision-making and process optimisation, you gain solutions that reduce manual work, accelerate funding decisions and improve control over financial processes.
We support banks, fintechs and supply chain finance providers in improving operational efficiency.
AI in credit and risk: faster decisions, fewer errors
In credit scoring, AI and machine learning models improve accuracy through advanced predictive analytics and a greater ability to tailor assessments to individual profiles, enabling extended lending, more effective risk evaluation, and lower default rates. In fraud detection, the same infrastructure supports real-time pattern recognition across transaction data, flagging anomalies before losses occur rather than after.
The shift underway is from reactive risk management to proactive, real-time decisioning, and financial institutions recognise that it only delivers consistent results when it runs on high-quality, integrated data. This is where the infrastructure question becomes directly relevant to outcomes. Models trained on clean, institution-owned data, with full lineage, consistent formatting, and no third-party data-sharing constraints, produce materially better results than those running on generic or fragmented inputs. Nearly 60% of financial services providers still report difficulty deploying and maintaining AI risk models, with data integration identified as the primary obstacle.
The operational efficiency gap between institutions that have solved their data management infrastructure and those that have not is already widening. Fraud detection and AML monitoring tend to deliver the fastest time-to-value among AI applications in finance, given their reliance on structured transaction data and clear success metrics. Credit and underwriting automation offers higher long-term ROI, but requires more foundational data work upfront, which is precisely why getting the infrastructure right before scaling these applications is a strategic decision.
Customer onboarding and regulatory reporting
According to Capgemini’s 2025 World Cloud Report, 59% of banks are deploying AI tools for customer interactions and client onboarding. The data insights here are straightforward: slow, manual onboarding contributes to 70% of banks losing clients, and the operational cost of KYC at scale is considerable. BCG’s 2025 global study found that banks applying AI to KYC are cutting costs by up to 50%, while simultaneously improving compliance coverage and reducing manual workload. AI chatbots can also easily enhance customer service with 24/7 support.
Regulatory reporting follows a similar logic. Compliance teams spend significant time on document search, version comparison, and evidence gathering; repetitive tasks that autonomous AI agents handle faster and with better traceability than manual processes. The more immediate value for most institutions is the reduction in reporting errors and the ability to respond to regulatory changes without manual rework across multiple systems.
The benefits of AI in finance and where institutions are still leaving value on the table
The benefits of AI in finance are not theoretical. Cost reduction, faster decisioning, stronger compliance coverage, and improved customer journeys are all measurable and documented across the sector. The gap that persists rests between early results and enterprise-wide delivery.
Where the returns are already visible
On cost, the direction is clear. AI automation reduces manual processing workloads, lowers error rates, and compresses the time required for high-volume, repetitive functions, with fraud prevention and compliance monitoring delivering the most immediate and measurable savings.
On decision speed, credit applications and financial operations that previously required days now resolve in minutes. Onboarding that once stretched across weeks now completes in a single session for standard cases. Customer experience follows: when back-office processes accelerate, the customer-facing result is faster answers, fewer errors, and less friction at points of high frustration.
Why those returns are not reaching scale
These are genuine, documented improvements. The problem is that most institutions are capturing them in pockets rather than across the organisation.
McKinsey’s State of AI 2025 found that while AI usage has risen sharply, only around one-third of organisations report scaling AI across the enterprise, with the message being that usage is up, but value at scale remains elusive. The reasons are consistent across institutions and markets. Eight in ten companies cite data limitations as a primary roadblock to scaling AI, with shaky data frequently identified as the core reason pilots do not translate into production systems that deliver reliable value.
In financial services, this plays out in specific and familiar ways. Data sits in systems that were never designed to talk to each other (core banking platforms, risk engines, CRM systems, compliance tools) each with its own data model, update cadence, and access controls. An AI model that performs well when trained on clean, consolidated data underperforms or fails when it encounters the reality of live institutional data environments. The problem is the infrastructure underneath the model.
The governance and ownership gap
Ownership compounds the difficulty. AI initiatives in banks often begin in a single function: fraud, credit, customer service; usually with a clear sponsor and a contained scope. Scaling across functions requires human oversight and coordination between teams with different priorities, different definitions of data quality, and different risk tolerances. McKinsey’s analysis identifies data fragmentation, workflow ambiguity, and governance gaps as the core scaling constraints, noting that budget alone does not resolve them, and that organisations can run pilots indefinitely without changing the way work actually gets done.
The institutions closing this gap share a common characteristic: they have treated data infrastructure and governance as prerequisites for AI investment, not as problems to solve later. The benefits of AI in finance are accessible but they accrue to institutions that have built the foundations capable of sustaining them at scale. For a growing number of banks, that conversation has moved from productivity to resilience.
Read more: AI infrastructure: a comprehensive guide to building your AI stack
Why resilience is replacing productivity as the primary driver
For most of the past decade, the case for AI in financial services was built on efficiency: faster processes, lower costs, reduced manual effort. That argument has not disappeared, but it is no longer the primary conversation in boardrooms and technology committees. The question has shifted from what artificial intelligence can deliver to whether institutions can sustain, govern, and control it at the scale they need.
This is a structural change in how the financial industry think about AI investment, and it is being driven by three converging pressures:
- regulatory compliance obligations that now reach directly into technology infrastructure,
- growing concerns about data sovereignty in an increasingly unstable geopolitical environment,
- and direct experience of what happens when AI systems built on cloud-dependent architectures encounter the limits of that dependency.
The regulatory floor has risen
DORA (the EU’s Digital Operational Resilience Act) entered into force in January 2025, applying unified ICT risk and resilience requirements to approximately 22,000 financial entities across the EU, with fines of up to 10% of annual turnover for serious breaches. The regulation imposes direct obligations around third-party risk management, data security, operational continuity, and the ability to demonstrate that critical functions can withstand and recover from technology disruptions, including those originating with external providers.
For financial institutions whose compliance screening, customer advisory, or document processing depends on an external AI API, an outage at the provider directly impacts the bank’s operational continuity, and that dependency now sits within DORA’s scope. The EU AI Act adds a further layer: AI systems used in credit scoring and financial decision-making are classified as high-risk under the EU AI Act, requiring documented data origins, model versioning, auditability, and explainability, with fines reaching €15 million or 3% of global turnover for non-compliance.
Sovereignty is becoming a strategic priority
Regulatory pressure is reinforced by a broader shift in how European financial institutions think about where their data sits and who controls it. An Accenture study published in late 2025 found that 62% of European organisations are seeking sovereign AI solutions in response to geopolitical uncertainty, with banking the leading sector, at 76%. The concern is not abstract: customer data, transaction histories, credit models, and risk assessments represent some of the most sensitive data any organisation holds, and the question of which jurisdiction governs its processing has become a material risk consideration.
With DORA in force and the EU AI Act close behind, European banks are under pressure to localise data and demonstrate regulatory control, but sovereignty alone does not resolve the governance and resilience requirements that regulators are demanding. The institutions navigating this most effectively are those treating infrastructure decisions and compliance obligations as a single problem rather than two separate workstreams.
Control and flexibility as competitive advantages
What emerges from this regulatory and geopolitical context is a reframing of on-premise AI infrastructure as a deliberate architectural response to specific operational requirements. An institution that implements AI on infrastructure it controls can include those systems in its own disaster recovery planning, test them under its own resilience framework, and operate them independently of third-party availability windows. It can demonstrate data residency and data privacy to regulators without relying on contractual assurances from cloud providers. It can adjust, retrain, and govern its models on its own terms.
Since January 2025, DORA has required financial institutions to actively manage risk and document cloud dependencies, and institutions that treat this as a compliance burden are missing its strategic dimension. The institutions that move first on infrastructure control are building an advantage that compounds: every AI capability added on a governed, resilient foundation is one that can scale without renegotiating the underlying architecture.
On-premise AI infrastructure: a strategic choice
For finance professionals operating under DORA, the EU AI Act, and GDPR simultaneously, running AI on infrastructure the institution owns and controls is increasingly the architecture that satisfies the most requirements with the fewest dependencies.
Data residency is verifiable rather than contractual, and regulators can audit it directly. Latency is predictable, which matters for real-time insights and applications in fraud detection and credit decisioning where response times affect both the customer experience and the reliability of the model output. Model behaviour is auditable end-to-end, with full visibility into training data, versioning, and decision logic. And capacity can be scaled on the institution’s own terms, without exposure to the pricing variability or availability constraints of shared cloud infrastructure.
None of this means cloud has no role. Most institutions operating at scale will run hybrid architectures, with on-premise infrastructure handling sensitive workloads and regulated processes while cloud environments support development, testing, or less constrained functions. The strategic question is which workloads require the level of control, auditability, and resilience that only institution-owned infrastructure can provide, and whether the organisation has the advisory and implementation capability to build and operate that infrastructure effectively.
This is where the gap between intent and execution tends to open. Defining the right architecture for a specific institution’s regulatory environment, data landscape, and AI roadmap requires both technical depth and familiarity with the compliance obligations that shape the design.
Future Processing's on-premise AI advisory and implementation offering is built around exactly this problem, helping financial institutions move from infrastructure decisions to working, governed AI systems without treating the two as separate projects.
A secure on-premise foundation for compliant AI development
Read the case studyWhat a scalable AI architecture looks like in practice
Getting AI to scale in a financial institution is not primarily a modelling problem. The models, in most cases, are available, proven, and well-understood. What determines whether they deliver value at enterprise level is the architecture surrounding them: the data pipelines feeding them, the systems they integrate with, the governance frameworks overseeing them, and the organisational conditions that allow them to be maintained and improved over time.
Integration as the foundation
The institutions that succeed at AI scale are those that integrate AI deeply within their core architecture rather than treating it as an auxiliary feature. In practice, this means connecting AI systems to the data sources that matter (core banking platforms, risk engines, CRM systems, document management) through stable, governed pipelines rather than one-off integrations built around individual projects.
Legacy infrastructure makes this harder than it sounds. Most core banking systems were not designed to expose data in the formats or at the frequencies that AI applications require. Building the integration layer that bridges these systems, without disrupting live operations or introducing new risk in maintaining compliance, is where many AI programmes stop developing. The technical work is tractable; the challenge is doing it in a way that holds up under audit, scales to additional use cases, and does not create fragile dependencies that surface as operational risk later.
Governance cannot be retrofitted
Only 12% of financial institutions have successfully implemented enterprise-wide AI strategies, and what separates them from the remaining 88% is not budget or board support, but execution discipline: robust data governance including data lineage and reproducibility, end-to-end platform engineering, and measurement capabilities that connect AI activity to actual business outcomes.
Governance is the component most consistently treated as something to address after deployment rather than before. In regulated environments, that sequence produces predictable problems: models go into production without documented lineage, decisions are made without auditable rationale, and when regulators ask questions, as they increasingly do, the answers require reconstructing records that were never systematically kept. Only 26% of organisations report having comprehensive AI security governance policies in place, a figure that reflects how far most institutions still are from the kind of mature governance that enterprise AI scale requires.
Building governance in from the start means defining data ownership before models are trained, establishing model monitoring before they go live, and setting clear accountability for what happens when outputs are challenged. It is more work upfront; it is also the only approach that does not generate larger remediation costs later.
The organisational conditions matter as much as the technical ones
Architecture decisions do not exist in isolation from the organisations that have to implement and sustain them. AI at scale requires cross-functional coordination between technology teams, risk and compliance functions, business units, and senior leadership that most financial institutions have not yet built into their operating models. Pilots succeed partly because they are contained: one team, one use case, one sponsor. Scaling requires the same rigour applied across functions that have different priorities and different definitions of success.
The institutions moving most effectively toward enterprise AI are those that have made explicit decisions about ownership: who is responsible for data quality, who governs model behaviour, who signs off on deployment into regulated processes. These are organisational questions as much as technical ones, and they need answers before the architecture can deliver what it promises.
The gap between where most financial institutions are today and where they need to be is a gap in foundations: data infrastructure, integration architecture, governance frameworks, and the organisational structures that make those foundations operational. Institutions that address the foundations now are building something that compounds: each AI capability deployed on solid infrastructure creates the conditions for the next one to scale faster and with less risk.
Improve your financial operations with AI and automation
From invoice intake to decision-making and process optimisation, you gain solutions that reduce manual work, accelerate funding decisions and improve control over financial processes.
We support banks, fintechs and supply chain finance providers in improving operational efficiency.
