The Human Side of Finance: Moving from AI Experimentation to Accountable Adoption

Financial services has moved quickly through the first wave of AI experimentation. Across banking, wealth, insurance, capital markets and fintech, teams have trialled copilots, explored generative AI, built early agents, and identified use cases across sales, service, operations, legal, compliance, cyber and technology.

The energy is real. But the conversation is changing.

The question is no longer simply: what can AI do? It is now: how do we adopt AI in a way that is valuable, governed, explainable, measurable and human-led?

It’s exactly the journey I’ve been tracking through the Human Side of Finance series.

This shift matters because financial services is not a normal technology adoption environment. The sector operates inside a dense web of accountability: conduct, prudential resilience, operational risk, data protection, financial crime, cyber security, outsourcing, model risk and senior management responsibility. AI does not remove that accountability. If anything, AI makes accountability more visible.

The regulatory direction of travel reinforces this. The UK approach to AI is principles-based, with emphasis on:

The FCA, PRA and Bank of England have supported a similar principles-based, outcomes-focused approach, and regulators expect firms to take ownership of AI risks and benefits rather than wait for prescriptive rules.

The implication for financial services leaders is clear: AI will increase, not reduce, regulatory accountability.

The first phase of AI in financial services was characterised by curiosity and controlled experimentation. Firms ran productivity pilots, gave selected teams access to copilots, tested document summarisation, explored customer service scenarios, experimented with knowledge retrieval and assessed where agentic AI might reduce manual effort.

That phase was important. It helped organisations build confidence, understand the art of the possible and identify where AI could help colleagues move faster. But experimentation often happens around the edge of the operating model. Adoption happens inside the operating model.

That is where the harder questions begin:

These are not technology questions alone. They are leadership, governance, literacy and accountability questions.

In FS, governance is sometimes framed as a brake on innovation. That is the wrong framing for AI. Good governance is what allows AI to scale.

Without governance, organisations get experiments, enthusiasm and isolated productivity gains. With governance, organisations get repeatable adoption, clearer ownership, reusable design patterns, better assurance and a credible route from proof of concept to production.

This becomes especially important as AI moves from individual productivity into agents and workflow automation. A colleague using AI to summarise a meeting is one thing. An agent that draws on enterprise data, interacts with a workflow, supports a regulated process or influences a customer or colleague outcome is another.

That is the difference between experimentation and adoption. Experimentation asks: can this use case work? Whereas adoption asks: can this use case be owned, governed, measured, supported and improved over time?

One of the most revealing things about AI adoption is that it quickly exposes how decisions really get made. During early experimentation, a motivated team can often move quickly. Moving into adoption requires wider alignment: business owners, risk, compliance, technology, data owners and senior leaders all need a shared view of value, feasibility and accountability.

In client work, this is where the human side of AI becomes visible. Some teams arrive with well-defined use cases and clear ownership. Others need help to clarify requirements, align internally and agree who will own the outcome after launch. This is not a failure of ambition; it’s a normal part of moving from exploration into operational change.

That advisory discipline matters. The strongest AI programmes are not the ones with the longest list of ideas. They are the ones with the clearest mechanism for deciding which ideas deserve to move forward – which takes me onto AI literacy

The next major human shift is AI literacy. In the early phase, many organisations treated AI literacy as enablement: training sessions, prompt-writing tips, demos and adoption communications. Those still matter. But in financial services, literacy must now be treated as part of the control environment.

People need to understand:

That literacy cannot be generic, and everyone needs it. Executives need decision literacy. Risk and compliance teams need assurance literacy. Technology teams need platform and control literacy. Business teams need workflow literacy. Frontline colleagues need safe-use literacy.

The goal is informed challenge rather than blind trust in AI.

AI adoption in financial services cannot rely on belief. It needs evidence.

Leaders need a disciplined way to connect AI initiatives to outcomes that matter: customer experience, colleague productivity, operational efficiency, risk reduction, quality improvement, speed to resolution, cost-to-serve, control effectiveness or revenue enablement.

This creates a new advisory opportunity in financial services: helping clients create a repeatable AI value discipline. Every priority use case should have an owner, a value hypothesis, a baseline, a success measure, a risk view and a decision on whether to scale, reshape or stop.

One of the more mature signs of AI advisory is knowing when not to build an AI agent.

In the current market, there is a natural tendency to turn every process challenge into an agent opportunity. But in financial services, the better answer may sometimes be a workflow change, a data-quality improvement, a standard platform capability, better use of CRM, or stronger governance around an existing tool.

The most credible advisory position is therefore not ‘AI everywhere’. It is AI where it creates measurable value and can be governed responsibly.

AI adoption is not just another application rollout. It changes how work is initiated, executed, reviewed, and improved. That requires an AI operating model.

An effective AI operating model connects strategy to delivery. It defines how opportunities are identified, how use cases are prioritised, how value is measured, how risk is assessed, how solutions are designed, how approvals work, how agents are monitored, and how colleagues are supported after launch.

A practical operating model should include:

This is where many firms now need help. They have enough ideas but need a repeatable way to move the right ideas into governed adoption.

The irony of AI adoption is that the more capable the technology becomes, the more important the human system becomes.

AI can help with analysis, summarisation, discovery, workflow support, content generation and execution. But people remain responsible for intent, context, challenge and outcomes.

Financial services firms will still need judgement, empathy, relationship management, ethical reasoning, commercial understanding and regulatory accountability.

The next phase of AI adoption will not be defined by experimentation alone. It will be defined by the firms that can combine innovation with governance, productivity with accountability, and automation with literacy.

The opportunity is significant. But in financial services, the path to value runs through trust. And trust is built by people!

Interested in learning more about the Human Side of Finance? Get in touch with our team today →