Why transforming a legacy bank requires more than technology

The conversation about artificial intelligence in banking is often dominated by technology: large language models, machine learning, intelligent automation, cloud platforms and data. Yet a bank does not become AI-powered simply because it buys new technology or gives employees access to a chatbot.

The real transformation happens when people use AI to redesign how the bank understands customers, makes decisions, manages risk and performs work.

This presents a particular challenge for legacy banks. Over many years, they have accumulated complex processes, fragmented systems, functional silos, approval layers and manual controls. These structures were often created for valid reasons, including growth, regulation and risk management. But they can make it difficult to respond quickly to changing customer expectations.

Adding AI to this environment without changing the operating model may only automate existing complexity. A slow, fragmented process does not become customer-centric merely because one task within it is performed faster.

Building an agile, AI-powered banking operation therefore requires more than data scientists and engineers. It requires employees throughout the bank to develop new non-technical capabilities while applying the banking knowledge they already possess.

What is changing in AI-powered banking operations?

Traditional banking operations have generally been designed around functions, products and queues of work. Employees receive applications, check documents, apply rules, investigate exceptions, capture information, respond to customers and pass cases from one department to another.

In an AI-powered operation, much of this work can be reorganised.

AI can read and classify documents, summarise customer histories, identify missing information, detect unusual patterns, predict potential outcomes, recommend a next action and draft customer communications. AI assistants can help employees retrieve policies, prepare reports and navigate complex cases. Routine decisions can be automated within defined thresholds, while unusual or high-risk cases are directed to people with the appropriate expertise.

This changes the nature of operational work in several ways:

● Employees move from processing every case to supervising automated workflows and resolving exceptions.

● Work is organised increasingly around end-to-end customer journeys rather than narrow departmental tasks.

● Decisions are made closer to the customer, supported by real-time information.

● Operational teams become more multidisciplinary, combining product, process, risk, data and customer expertise.

● Continuous experimentation and improvement replace occasional large-scale process redesigns.

● Performance is measured through customer and business outcomes, not merely volumes processed or tasks completed.

● Managers spend less time allocating work and more time coaching people, removing obstacles and improving systems.

AI can make the operation faster and more responsive. It can also expose unnecessary hand-offs, duplicated controls and policies that have become disconnected from the customer need they were originally intended to address.

However, none of this happens automatically. Technology can accelerate a well-designed operation, but it can also scale poor decisions, unfair treatment and operational weaknesses.

What does not change?

Banking remains a business of trust.

Customers still expect the bank to protect their money and personal information, treat them fairly, remain available when needed and take responsibility when something goes wrong. Regulators, shareholders and society still expect sound governance, financial stability and accountable decision-making.

AI cannot assume legal, ethical or fiduciary accountability. It cannot be held responsible for an unfair credit decision, an incorrectly blocked transaction, a poorly handled vulnerable customer or the disclosure of confidential information. Accountability remains with the bank and the people who design, approve, use and supervise its systems.

Technology also cannot replace the full value of human context. A model may detect a pattern, but an experienced employee may recognise that a customer has been retrenched, is experiencing financial abuse or requires an appropriate form of forbearance. An AI assistant may retrieve the relevant policy, but a person must still exercise judgment when policies conflict, the facts are incomplete or the consequences are significant.

The enduring requirements of banking include:

● Sound professional judgment

● Customer empathy and fairness

● Ethical courage

● Domain and product knowledge

● Risk awareness and control discipline

● Protection of confidential information

● Clear accountability

● The ability to explain and defend decisions

● Constructive challenge and escalation

● Trust-based human relationships

AI changes how these capabilities are applied; it does not make them obsolete.

The non-technical capabilities an AI-powered bank needs

1. AI literacy

Every employee does not need to know how to build a model. Every employee does need to understand, at an appropriate level, what AI can do, what it cannot do and how it can fail.

AI literacy includes understanding that outputs are probabilistic, that fluent language is not proof of accuracy, and that results can be affected by poor data, bias, outdated information or badly framed instructions. Employees need to recognise when an output should be verified, challenged, escalated or rejected.

2. Critical thinking and professional scepticism

An AI-generated answer can sound authoritative even when it is wrong. Employees must distinguish evidence from inference, identify missing context, test assumptions and compare recommendations with approved policies and reliable information.

The right relationship is neither blind trust nor blanket resistance. AI should be treated as a capable assistant whose work must be evaluated in proportion to the risk of the decision.

3. Problem framing

One of the most valuable skills in an AI-enabled organisation is defining the right problem. Before proposing an AI solution, employees should ask:

● What customer or operational problem are we solving?

● What is causing it?

● Is AI the appropriate response?

● Which decision or task needs to improve?

● How will success be measured?

● What new risks could be introduced?

Poorly framed problems create sophisticated solutions that deliver little value.

4. Customer empathy and ethical judgment

AI can optimise an outcome without understanding whether that outcome is humane, fair or appropriate. Employees must consider the customer’s circumstances, vulnerability and ability to challenge a decision.

This is especially important in lending, collections, fraud management, complaints and account closures. Operational efficiency must never become an excuse for impersonal or unjust treatment.

5. Data literacy

Employees need to understand the information behind AI-supported decisions. This includes interpreting basic probabilities and trends, recognising incomplete or biased data, questioning the source and freshness of information and distinguishing correlation from causation.

Data literacy also means knowing which information may be entered into an AI tool and protecting personal, commercially sensitive and confidential data.

6. Process and systems thinking

Legacy organisations often optimise individual departments while making the complete customer journey more difficult. Employees need to see how a decision in one part of the bank affects downstream teams, risk, cost and the customer experience.

This capability helps teams decide which work should be automated, which should be augmented and which should be removed entirely.

7. Human–AI workflow design

Employees closest to the work should help define the division of labour between people and machines. They understand exceptions, informal workarounds and customer realities that may not appear in formal process documentation.

They should help decide:

● What AI may recommend

● What it may execute automatically

● When human approval is required

● What should trigger an escalation

● How errors will be detected and corrected

● How customers can question an outcome

● Who remains accountable

8. Collaboration across traditional boundaries

AI-powered operations require product, operations, technology, data, risk, compliance, legal and customer teams to work together. Sequential hand-offs and late-stage approvals slow delivery and frequently produce rework.

The necessary skills include listening, translating between disciplines, constructive disagreement, shared accountability and the ability to balance customer experience, commercial value and risk.

9. Experimentation and learning agility

AI use cases should be treated as hypotheses to be tested, not technology projects whose success is declared at launch. Employees need to define measurable outcomes, test solutions with users, examine unintended consequences and improve or stop initiatives based on evidence.

10. Communication and explanation

Employees must be able to explain AI-supported decisions to customers, colleagues, executives, auditors and regulators in plain language. “The system decided” is not a meaningful explanation.

Clear communication also includes writing effective instructions for AI, recording assumptions, documenting exceptions and giving precise feedback to technical teams.

11. Change resilience and personal adaptability

AI may remove tasks that have historically defined a person’s role or status. Employees will need to release low-value manual work, learn continuously and shift from processing to judgment, supervision, improvement and relationship management.

Leaders must approach this transition honestly. They should explain how roles are changing, invest in reskilling and involve employees in redesigning their work. Fear cannot be managed through slogans about innovation.

Turning existing experience into AI-enabled capability

Experienced banking employees should not assume that their knowledge has suddenly lost its value. Domain expertise is one of the most important foundations of responsible AI adoption.

A credit specialist knows which affordability indicators deserve scrutiny. A fraud investigator recognises emerging criminal behaviour. A complaints specialist understands how apparently minor process failures affect customers. An operations employee knows where work is duplicated and which exceptions routinely defeat the standard process. A relationship manager understands the context that data alone may not reveal.

These employees are needed to train, test, challenge and improve AI-enabled operations.

However, experience without AI literacy will become progressively less sufficient. Employees who cannot use AI tools, evaluate their outputs or contribute to redesigned workflows may find that the transactional portion of their roles diminishes while new opportunities go to colleagues who combine domain knowledge with AI-enabled ways of working.

The career risk is not simply that AI will replace people. It is that people who can work effectively with AI will increasingly perform roles previously performed by people who cannot.

The most valuable future employee will therefore combine three forms of capability:

1. Deep knowledge of banking, customers or operations

2. Human judgment, empathy and ethical responsibility

3. The ability to use, supervise and improve AI-enabled systems

High-level learning journeys for bank employees

The learning journey should not be identical for everyone. It should begin with a common foundation and then develop according to the employee’s role.

Journey 1: AI-aware colleague

Who it is for: Every employee

Learning outcomes:

● Understand basic AI concepts and common banking applications

● Recognise hallucination, bias, privacy and security risks

● Use approved AI tools safely

● Write clear prompts and provide appropriate context

● Verify outputs against reliable sources

● Know when and how to escalate concerns

Practical application: Use an approved assistant to summarise non-sensitive material, retrieve a policy or prepare a first draft, followed by human verification.

Journey 2: AI-enabled domain practitioner

Who it is for: Operations, credit, fraud, payments, servicing, finance, compliance and other domain specialists

Learning outcomes:

● Apply AI to role-specific tasks and decisions

● Evaluate outputs using domain expertise

● Identify suitable use cases and operational risks

● Interpret basic confidence measures and performance indicators

● Handle exceptions and maintain meaningful human oversight

● Provide structured feedback to improve the solution

Practical application: Compare an AI-generated case summary or recommendation with source information, identify omissions and document the appropriate action.

Journey 3: Human–AI process designer

Who it is for: Process owners, business analysts, product managers, change practitioners and operational leaders

Learning outcomes:

● Map end-to-end customer journeys and decision points

● Identify tasks to eliminate, automate, augment or retain

● Define human approvals, thresholds and escalation routes

● Design customer recourse and error-correction mechanisms

● Establish outcome, risk and adoption measures

● Run controlled pilots and incorporate user feedback

Practical application: Redesign an onboarding, fraud or complaints journey and test it with frontline employees and customers before scaling.

Journey 4: AI risk and governance practitioner

Who it is for: Risk, compliance, legal, audit, privacy, model-risk and control professionals

Learning outcomes:

● Understand AI-specific conduct, model, privacy and security risks

● Assess data lineage, fairness, explainability and accountability

● Design controls proportionate to the consequence of the use case

● Challenge AI solutions constructively and early

● Monitor outcomes across different customer groups

● Evaluate incidents, overrides and customer appeals

Practical application: Complete a risk assessment for an AI-supported credit or fraud decision, including monitoring and human-override requirements.

Journey 5: AI-enabled leader

Who it is for: Team leaders, executives and board members

Learning outcomes:

● Connect AI investment to customer, risk and commercial outcomes

● Prioritise use cases rather than chase technology trends

● Establish clear decision rights and accountability

● Lead multidisciplinary, empowered teams

● Create psychological safety for employees to challenge AI

● Govern workforce transitions responsibly

● Measure realised value, not only deployment activity

Practical application: Sponsor a value-stream transformation with clear customer outcomes, risk boundaries, workforce implications and benefits tracking.

Journey 6: AI product and transformation specialist

Who it is for: Employees seeking deeper careers in AI-enabled banking transformation

Learning outcomes:

● Translate business problems into product requirements

● Understand the AI lifecycle without necessarily becoming a programmer

● Coordinate business, data, technology and control specialists

● Design experiments and evaluate solution performance

● Manage adoption and operating-model change

● Build a portfolio of demonstrated AI-enabled improvements

Practical application: Lead a multidisciplinary team from problem discovery through pilot, controlled deployment and benefit realisation.

Learning must happen through work

Awareness courses alone will not transform the bank. Learning should combine short conceptual modules with practical application, coaching and evidence of competence.

A useful progression is:

1. Learn: Understand the concept and its banking relevance.

2. Practise: Use an approved tool in a safe environment.

3. Apply: Solve a real operational or customer problem.

4. Demonstrate: Show that the output is accurate, safe and valuable.

5. Teach: Share the learning with colleagues and improve the standard practice.

Banks can support this through role-based academies, simulations, multidisciplinary improvement projects, communities of practice, reverse mentoring and internal accreditation. Performance objectives should reward responsible experimentation, knowledge sharing and measurable customer outcomes.

The leadership challenge

Legacy-bank transformation will fail if leaders treat AI as a cost-reduction programme delegated to technology teams.

Leaders must be willing to simplify structures, remove redundant approvals and give multidisciplinary teams authority to improve complete customer journeys. At the same time, they must preserve independent challenge, regulatory accountability and robust controls.

They must also guard against using AI primarily to intensify work or monitor employees. Trust will deteriorate if colleagues experience AI as something being done to them rather than a capability being built with them.

The leaders who are most likely to succeed will involve experienced employees in designing the new operation, confront the implications for roles honestly and create credible routes for people to develop.

The future belongs to the augmented banker

An AI-powered bank is not a bank without people. It is a bank in which people spend less time searching for information, capturing repetitive data and moving work between queues—and more time exercising judgment, improving processes, solving customer problems and managing exceptions.

Existing banking skills remain enormously valuable. Domain knowledge, relationships, professional judgment, empathy and risk discipline cannot simply be downloaded into an organisation. But these capabilities must now be combined with AI literacy.

For employees, the message is both urgent and hopeful: do not discard what you know. Build on it. Learn how AI works, use it responsibly, challenge it intelligently and help redesign the work around it.

The employees most at risk of redundancy will not necessarily be those without technical qualifications. They will be those whose contribution remains limited to repeatable tasks and who do not develop the ability to work alongside intelligent systems.

The employees with the strongest prospects will be those who can bring together human wisdom and machine capability—using technology to make banking faster and simpler without surrendering the judgment, fairness and accountability on which trust depends.