AI and the Financial Services Sector

The value case is no longer the issue

The financial services AI value pool is material. McKinsey has estimated that generative AI could add US$200 billion to US$340 billion in annual value across banking. The practical implication for executives is that AI is no longer a marginal productivity tool. It is now relevant to the economics of the institution.

The strongest use cases are not abstract. They sit close to decisions and workflows that already carry commercial weight.

Fraud detection can reduce losses, lower false positives and improve customer trust at the point of transaction. Financial crime and AML use cases can reduce case handling effort, improve triage and allow specialist teams to focus on higher risk activity. Credit decisioning can improve approval speed, risk selection and customer access. Contact centre AI can reduce after call work, improve service consistency and return capacity to frontline teams.

Across Codex’s work in market, this is where executive interest is strongest: use cases where AI can be tied to a measurable business outcome, not simply deployed because the technology is available.
The issue is that many institutions are still capturing only a fraction of the expected value. McKinsey has found that large companies have captured, on average, only 31 percent of expected revenue uplift and 25 percent of expected cost savings from digital and AI transformations. For financial services leaders, the implication is clear. The value case may be approved, but value realisation requires a different level of execution discipline.

The gap between the pilot and the P&L

In financial services, AI pilots often succeed technically before they fail organisationally.
A model can perform well in a controlled environment, but still fail to scale if the data is hard to access, the workflow is not redesigned, the risk controls are unclear, or the business owner is not accountable for realised benefits. This is where return gets stuck.

What we see in production environments is that AI value depends less on the number of use cases and more on the institution’s ability to move proven use cases into the operating rhythm of the business. That means embedding AI into lending processes, fraud operations, compliance workflows, service centres, engineering teams and management reporting, with clear accountability for the outcome.

For Australian financial services organisations, this matters because the operating environment is already complex. Customer expectations are rising. Fraud and scam activity is more sophisticated. Regulatory scrutiny remains high. Productivity pressure is constant. Technology teams are managing cloud modernisation, cyber resilience, data quality and legacy remediation at the same time.

AI cannot sit outside that reality. It has to be designed into it.

The 77 percent problem

One of the strongest market signals is that 77 percent of banking innovation leaders cite legacy systems restricting data availability as the primary barrier to AI adoption. The same proportion point to poor data quality, and 71 percent cite difficulty accessing real time data.

For executives, this reframes the AI agenda. The AI strategy is not separate from the data strategy. It is now one of the strongest tests of whether the data strategy is working.

A fraud model needs trusted transaction, customer, device, behavioural and case data. A credit decisioning model needs current, consistent and explainable data across internal and external sources. A service assistant needs governed access to product, policy, customer and interaction history. An AI agent cannot be allowed to act across systems unless the organisation understands the data, controls the permissions and monitors the outcomes.

This is why cloud modernisation and data readiness should be treated as business value enablers, not technology projects. Modern data platforms, governed data products, common definitions, real time integration and secure cloud foundations reduce the distance between a validated use case and production value.

The institutions capturing AI value are not necessarily those with the largest innovation portfolio. They are the ones that made the earlier commitment to data foundations, platform reuse and governance.

Governance is a scale mechanism

Financial services leaders often worry that governance will slow AI down. In practice, weak governance slows AI down more.

Without clear governance, pilots remain contained because risk, compliance, legal, cyber and business teams are not confident enough to scale them. Controls are added late. Assurance becomes manual. Ownership is unclear. Benefits are debated rather than measured. The organisation becomes busy, but not materially better.

Good governance changes that. It defines which AI use cases can be used, under what conditions, with what level of human oversight, and with which monitoring and escalation paths. It creates the confidence to move from experimentation into trusted enterprise use.

This becomes more important as AI shifts from assisting people to taking action within defined boundaries. Agentic AI increases the urgency because the risk profile changes. The concern is no longer only that an AI system may produce the wrong answer. It is that an AI enabled workflow may take the wrong action.

Current regulatory guidance is still catching up with this shift. Australian financial services organisations should not wait for regulation to resolve every detail. They need to establish their own practical controls now, aligned to local privacy, data, security, accountability and operational resilience expectations.

The executive question should be direct: which decisions are we prepared for AI to recommend, influence or execute, and what controls must be in place before that happens?

CEO sponsorship changes the return profile

AI at scale is not a technology team agenda. It crosses product, risk, operations, finance, data, technology, legal, compliance and the frontline. That is why executive sponsorship matters.

CEO sponsored AI delivers 2.5 times more value than efforts initiated further down the organisation. The implication is not that every AI decision must sit with the CEO. It is that AI value requires executive alignment on priorities, funding, operating model, risk appetite and benefit ownership.

For financial services organisations, the unresolved questions are often practical.

• Who owns the value case after the pilot?
• Who funds the integration work?
• Who signs off on acceptable risk?
• Who changes the workflow?
• Who measures whether capacity, cost, revenue or risk outcomes actually improved?

If these questions are not answered early, AI initiatives drift. They remain impressive demonstrations rather than durable business capability.

A practical agenda for financial services leaders

The next phase of AI in financial services should be more disciplined, not more experimental.

First, focus investment on value pools that matter to the institution. Fraud loss reduction, AML productivity, credit cycle time, customer service efficiency, claims processing, engineering productivity and regulatory reporting effort are all areas where the link to business value can be made explicit.

Second, fix the data foundations that constrain scale. This includes governed data products, real time access where needed, consistent definitions, lineage, quality controls and secure cloud architecture. The goal is not perfect data. The goal is usable, trusted data for priority decisions.

Third, build governance into delivery. AI risk, privacy, security, model validation, compliance and business ownership need to be designed into the delivery model from the start. Late stage assurance is expensive and slows adoption.

Fourth, use platforms and patterns that can be reused. Each AI use case should make the next one easier by contributing to shared architecture, reusable controls, common integration patterns and tested delivery playbooks.

Fifth, measure return properly. Number of pilots, number of users and number of models are activity metrics. Executives need to see fraud losses reduced, handling time improved, approval speed increased, compliance effort lowered, customer friction reduced or employee capacity returned to higher value work.

The executive provocation

The Return on AI gap in financial services will not be closed by running more pilots. It will be closed by institutions that modernise the data estate, build governance into the platform layer and redesign the operating model around measurable value.

The value is proven. The barriers are known. The institutions that move first on data foundations and governance will capture a disproportionate share.

The question for executive teams is simple: are you investing in more experiments, or in the operating model required to scale the ones that already work?

Read more of Adrian’s ‘Return on AI’ series here.