Introduction
Artificial intelligence has become a significant investment area for banks, with applications ranging from fraud detection and customer service to compliance, risk management and operational automation. Yet demonstrating that an AI system works in a controlled pilot is only the beginning of the adoption process.
The more difficult challenge is integrating those systems into live banking operations while maintaining appropriate governance, accountability and human oversight.
The World Economic Forum forecasts global spending on AI in banking will reach $97 billion by 2027, highlighting the scale of investment entering the sector. But increasing expenditure does not necessarily translate into enterprise-wide adoption.
Ben Saunders, co-founder of AI consultancy WeBuild-AI, argues that the obstacle to scaling AI is often less about the underlying technology or regulation and more about organisational readiness. In his view, banks need operating models that allow AI systems to be governed, monitored and trusted before experimental projects can become core business capabilities.
Why AI Pilots Often Struggle to Reach Production
AI systems can perform effectively within controlled environments because the scope of the problem, available data and operating conditions are relatively well defined.
A production banking environment is considerably more complicated.
An AI application may need to interact with legacy systems, customer data, compliance processes and existing workflows while operating under defined risk controls. Responsibility for the system may also be shared between technology, business, risk and compliance teams.
This creates a gap between technical proof and operational deployment.
A successful pilot can demonstrate that an AI model is capable of performing a particular task. It does not necessarily demonstrate that an organisation has the infrastructure, governance or workforce processes required to operate that system at scale.
Without those foundations, banks risk repeatedly launching proof-of-concept projects without establishing a path for successful applications to become permanent capabilities.
AI Adoption Is an Operating-Model Challenge
Treating AI primarily as a technology project can make it difficult to establish who owns the resulting system and how its performance should be measured.
A scalable AI programme requires business leaders to understand what a system is intended to achieve, how its performance will be evaluated and who remains accountable for its outputs.
That can involve several functions:
- Technology teams responsible for infrastructure and integration.
- Compliance teams assessing regulatory requirements.
- Risk teams evaluating potential operational and financial exposure.
- Business teams defining the use case and expected outcomes.
- Senior leadership determining whether the investment delivers sufficient value.
When these responsibilities are established early, AI adoption becomes part of the organisation’s operating model rather than an isolated technology experiment.
Trust Is More Than Model Accuracy
Trust in banking AI is not simply a question of whether a model produces accurate results.
Financial institutions also need confidence that AI systems can be monitored, audited and governed appropriately.
Questions may include:
- What data does the system use?
- Who owns that data?
- How are AI-generated decisions reviewed?
- Who is responsible when the system produces an incorrect result?
- Can the organisation reconstruct how an important output was produced?
- When must a human intervene?
- How is model performance monitored after deployment?
These questions become increasingly important as AI moves closer to consequential decisions.
A system used to summarise documents presents a different risk profile from one involved in credit decisions, fraud investigations or regulatory processes.
The appropriate governance structure therefore needs to reflect the use case rather than applying a single approach to every AI deployment.
Connecting AI Investment to Business Value
Another barrier to scaling AI is the difficulty of translating technical performance into measurable business outcomes.
Leadership teams need to understand why a particular AI application deserves continued investment.
Potential measures can include:
- Reduced processing time.
- Lower administrative workload.
- Faster decision-making.
- Improved operational resilience.
- Greater capacity for compliance teams.
- Reduced manual processing.
For example, AI can support KYC and AML operations by helping draft or review regulatory documentation. The potential benefit is not necessarily eliminating compliance professionals, but allowing them to spend less time on repetitive document-related work and more time on activities requiring professional judgement.
A clearly defined value case can also help different departments understand why an AI system is being introduced and what role it will play.
Governance Needs to Be Designed Into AI Projects
One of the most important differences between experimentation and production is when governance enters the development process.
If risk and compliance teams become involved only after a successful pilot has been completed, organisations may discover that the system cannot easily satisfy the requirements needed for deployment.
Embedding those stakeholders earlier can help address issues such as data governance, accountability, monitoring and human oversight during development.
This approach does not necessarily eliminate regulatory complexity. Instead, it can reduce the risk of discovering fundamental governance problems after significant resources have already been invested in a project.
Building Repeatable AI Frameworks
A bank that successfully deploys one AI system still faces the question of how to repeat that success.
A collection of isolated projects can create fragmented technology, inconsistent controls and different approaches to monitoring.
A more scalable approach involves developing repeatable processes for evaluating and deploying AI.
Such a framework could establish common principles for:
Use-Case Selection
Banks can assess whether a proposed application has a clear business problem, measurable benefit and manageable risk profile.
Data Governance
Teams can establish what information the AI system can access, how that data is managed and who is responsible for it.
Risk Assessment
Potential operational, compliance and model risks can be evaluated before deployment.
Human Oversight
Organisations can define which outputs require human review and which activities can be automated.
Performance Monitoring
AI systems need to be monitored after deployment to determine whether performance remains within expected parameters.
A repeatable framework can allow successful applications to scale without requiring every project to develop an entirely new governance structure.
Human Oversight Remains Important
Increasing automation does not remove the need for human decision-making.
In banking, there are circumstances where an AI system can efficiently process information but should not independently determine the final outcome.
Human oversight can be particularly important when decisions involve customers, regulatory obligations or material financial consequences.
The objective is therefore not necessarily maximum automation. It is to determine where automation provides value while maintaining appropriate control over decisions that require human judgement.
This also affects employee adoption. Staff are more likely to integrate AI into existing workflows when they understand what the system does, what its limitations are and when they are expected to intervene.
Operational Resilience and AI
AI adoption can also influence operational resilience.
If an AI system becomes embedded in a critical banking process, the organisation must consider what happens if the system becomes unavailable, produces degraded results or requires intervention.
This creates another reason to treat AI as part of the operating model.
Banks need to understand how AI-dependent workflows connect to existing systems and what alternative processes are available when automated services cannot operate as expected.
A pilot may never encounter those conditions. A production system inevitably will.
The Next Phase of Banking AI
The next stage of AI adoption in financial services is likely to depend less on the number of experiments a bank can conduct and more on how effectively it can industrialise successful applications.
That means establishing ownership, integrating AI into existing workflows, creating appropriate governance and developing measurable performance criteria.
The distinction is important because AI experimentation and AI transformation are fundamentally different activities.
Experimentation asks whether a technology can perform a task.
Transformation asks whether the organisation can operate that technology reliably, govern it appropriately and generate measurable value from it over time.
Outlook
Banking institutions are unlikely to reduce their interest in AI as investment continues to grow and new applications emerge.
The challenge will be converting that investment into sustainable operational capabilities.
Banks that build repeatable governance and implementation frameworks may be better positioned to move successful AI applications from isolated pilots into broader deployments. Those that focus primarily on experimentation could continue accumulating proofs of concept without achieving comparable enterprise-wide impact.
Trust will remain central to that transition. Employees, customers, compliance teams and executives need confidence not only in what an AI system can do, but also in how it is controlled when it becomes part of a critical financial process.
Conclusion
The future of AI in banking will depend on more than increasingly capable models.
Banks need the organisational infrastructure to deploy those systems responsibly, including clear ownership, data governance, risk controls, human oversight and measurable business objectives.
The transition from pilot to production is therefore an operating-model challenge as much as a technological one. AI systems that can demonstrate both performance and accountable governance are more likely to become lasting components of banking operations.
For financial institutions, the defining question is no longer simply whether AI works. It is whether the organisation is prepared to trust, govern and operate it at scale.
