Key Takeaways
- The financial services industry is entering a new phase of artificial intelligence (“A.I.”) adoption, with global financial institutions (“FIs”) investing an estimated USD 76 billion in A.I.-related initiatives in 2025, a number we forecast to reach USD 132 billion by 2030.
- Despite vast investments and institutional ambition, most financial institutions struggle to translate A.I. initiatives into measurable business value and clear returns on investment. In many cases, deployment activity has outpaced the development of the organisationalfoundations required to support it. Strategy is frequently backward engineered around available technology, risk is considered only after design and development have progressed, and existing operating models are not equipped to govern, implement, and scale A.I. effectively.
- To overcome these barriers, financial institutions should adopt an integrated roadmap across three interconnected layers:
- Strategise: A.I. should be treated as an enabling capability to deliver the enterprise strategy rather than as a standalone technology agenda. Institutions should balance outside-in awareness of emerging technological possibilities with an inside-out assessment of strategic priorities, customer journeys, and operational workflows. This requires systematically identifying friction points across existing processes, determining where A.I. can generate meaningful value, and applying non-negotiable eligibility thresholds before prioritising viable use cases. Beyond improvements to existing processes, there is a broader opportunity for institutions to fundamentally reimagine workflows around A.I. capabilities, which can potentially unlock new ways of delivering value.
- Govern: Governance should be established before material resources are committed. Traditional approaches that treat governance as a downstream approval process are proving insufficient for dynamic A.I. environments. Financial institutions should implement systematic eligibility gates to exclude use cases that do not meet minimum viable requirements. Governance must also extend beyond use-case approval into the underlying workflow, defining how A.I. operates within controlled boundaries through runtime controls, human checkpoints, and more. When designed effectively, early governance can serve not only as a compliance safeguard, but also as a competitive edge for more effective and confident deployment by establishing clear boundaries.
- Enable: Once strategic priorities and permissible guardrails have been established, institutions must develop the technological and operational capabilities required for execution. From a technology perspective, this includes selecting an appropriate model architecture and developing the surrounding A.I. engineering capabilities, including prompt engineering, context management, and harnesses to deliver reliable, secure, and cost-effective outcomes. Enablement also requires a fundamental shift in people, roles, and ways of working. As A.I. becomes embedded across institutional workflows, frontline staff will increasingly direct A.I. outputs, middle managers will orchestrate mixed teams of humans and A.I. agents, and senior leaders will be responsible for setting strategic direction, boundaries, and organisational design. A.I. transformation therefore cannot rely solely on a traditional top-down technology rollout as meaningful value is often identified closer to the frontline, where real domain knowledge and operational realities are most visible.
- Ultimately, the quality of transformation will depend on how effectively institutions codify human expertise into the context, rules, controls, and validation logic that guide A.I.-enabled workflows. Technology can automate and augment execution, but it cannot independently determine how an institution should operate, which risks it should accept, or what constitutes an appropriate outcome. Those decisions must remain grounded in institutional strategy, domain expertise, and clearly assigned accountability.
- As financial institutions continue to advance their A.I. ambitions, particularly towards more autonomous and agentic models, they should conduct a structured readiness diagnostic covering strategy, governance, technology, people, and processes. This diagnostic can identify critical gaps, prioritise interventions, and establish a practical 12 to 24-month roadmap for scalable adoption. By strengthening these foundations before accelerating deployment, institutions can shift from fragmented experimentation towards controlled, repeatable, and measurable value creation.
