Adoption starts with the bank we already have
Most banks will first use AI to improve the bank they already have.
That is understandable. Copilots can help people find information, prepare analysis and produce routine material more quickly. Automation can remove repetitive work. Existing teams can often show benefits without changing the wider organisation.
The risk is that they settle an important design question too early. The current bank becomes the fixed part of the equation, and AI becomes something to insert into it.
The same departments remain in place. Work moves through the same handoffs. Decision rights stay where they are. Controls continue to sit around processes designed for older systems and much more limited access to intelligence.
AI is asked to perform inside yesterday's operating model.
The bank may become more productive while leaving its underlying shape largely untouched.
A bank can automate a great deal of work without challenging why that work exists.
Start with the obligation, not the existing process
Starting from first principles can sound unrealistic in a regulated bank. No bank has the freedom to ignore its history, customers, technology estate or regulatory obligations.
But first-principles work does not require a blank sheet.
It requires leaders to separate what the bank must preserve from what it has simply become accustomed to.
Capital, conduct, privacy, resilience, model risk, record keeping and accountability are genuine constraints. A particular committee, queue, reconciliation or sequence of handoffs may not be.
Many operating processes were designed around the limits of the systems available at the time. Information had to be assembled manually. Expertise sat in specialist teams because it could not easily travel. Controls were applied at fixed checkpoints because continuous oversight was not practical.
Better intelligence changes some of those conditions.
Take a lending decision. If an AI-enabled process can assemble the relevant evidence, test it against policy, identify missing information, surface exceptions and prepare a reasoned recommendation, the bank should not assume the same case still needs to move through the same sequence of teams.
The design work begins with more direct questions:
- Which decisions require human judgement or formal accountability?
- Which controls can run continuously rather than at a later checkpoint?
- What evidence must be retained, and how can it be captured once?
- Where does a customer need empathy, explanation or discretion?
- What should trigger escalation, pause or review?
Those questions lead well beyond the choice of a copilot.
Guardrails make ambition usable
Regulation is sometimes presented as the reason banks cannot move as far or as quickly with AI.
In practice, well-designed guardrails make a more ambitious model possible.
They define what a system may do, the data it may use, the evidence it must produce and the conditions under which a person must intervene. They make authority explicit and give Risk, Compliance and Internal Audit something concrete to assess.
Poorly designed constraints can certainly slow progress. Vague policies, duplicated approvals and blanket prohibitions usually do.
Good constraints force the bank to be precise about decision rights, risk appetite, explainability, monitoring and accountability. That makes AI safer to operate and easier to scale.
Control still has to hold when intelligence is distributed differently across the bank.
This belongs with the leadership team
AI is often given to a technology, data or innovation function to progress. Those teams have an important role, but they cannot decide the future shape of the bank on behalf of everyone else.
Once AI begins to change how work is allocated, how decisions are prepared, where controls operate and when people intervene, the discussion is no longer mainly about tooling. It touches the operating model, accountability, workforce, architecture and customer proposition at the same time.
The leadership team has to explore that future together.
Business leaders need to decide where the customer and economic value sits. Risk and Compliance need to shape usable boundaries. Operations needs to work through exception handling. Technology and Data need to expose what is possible and what the current estate will support. HR needs to understand how roles and skills will change.
Delegating AI while keeping those decisions separate almost guarantees that the technology will be fitted around the existing organisation.
Prototypes make the choices real
It is difficult to redesign an organisation through abstract discussion.
Prototypes help because they give leaders something concrete to inspect.
A working version of a redesigned customer or operating journey exposes where the data is weak, where policy is ambiguous, where human judgement remains essential and which handoffs no longer add value. It also gives control teams a better basis for challenge than a general statement of AI ambition.
The prototype does not need to prove that the entire model is ready to scale. Its first job is to improve the decisions around it.
Building Henova, an AI-native health company, has sharpened this view for me. When AI is part of the design from the start, questions about evidence, consent, trust, escalation and human responsibility arrive much earlier. They are not governance added after the product. They shape the product and operating model together.
Banks will have more inherited complexity to work through, but the design discipline is the same.
Keep people where people matter
None of this removes the human role. It should make that role more deliberate.
People remain essential where judgement, trust, empathy, accountability and leadership carry weight. They should decide in genuinely ambiguous situations, support customers through consequential moments, challenge outcomes and remain answerable for the system.
They should spend less time searching across fragmented systems, re-keying information, preparing standard material or acting as the connection between teams that cannot share context.
Design one journey differently
Banks should adopt useful AI tools. Early productivity gains will build capability, improve confidence and help teams understand the technology.
They should not let those gains define the full ambition.
A leadership team can start by taking one important customer or operating journey and redesigning it on the assumption that better intelligence is already available.
Keep the obligations. Keep the accountability. Keep the moments where human judgement and trust count.
Then challenge the accumulated process around them.
The result may require different roles, fewer handoffs, controls that operate earlier or continuously, and clearer interfaces between people and systems. A prototype will show where that is credible and where the bank is not yet ready.
That conversation is more demanding than an AI adoption plan. It may also stop the bank spending the next few years placing new tools into an operating model it would no longer choose to build.