The proliferation of generative and agentic AI has opened up significant opportunities for banks, insurers and financial services organisations.
AI-enabled customer experiences can surface products and services tailored to an individual's circumstances. AI-assisted transaction monitoring can identify potential fraud and money laundering more efficiently than traditional rules-based approaches. Agentic underwriting processes have the potential to free specialists from repetitive administrative work, allowing them to focus on complex decisions where human expertise delivers the greatest value.
The prize extends well beyond operational efficiency.
Organisations are beginning to see a path towards dramatically reducing the time between customer acquisition, product recommendation, approval and onboarding. Processes that once required multiple handoffs, manual reviews and days of waiting are increasingly becoming candidates for straight-through processing.
Banks already know this.
The industry has largely moved beyond asking whether AI can deliver value. Executive sponsorship exists. Funding is increasing. AI initiatives are steadily moving from controlled pilots into production environments.
The question is no longer whether AI can create value. The question is whether existing technology foundations can support AI at enterprise scale.
AI is exposing a problem that already existed
Many organisations are achieving promising outcomes through AI pilots and targeted production deployments. The challenge emerges when they attempt to move beyond isolated use cases and scale AI across the wider enterprise.
At that point, a common reality surfaces. Most banks are carrying decades of accumulated technology decisions. Core banking platforms, lending systems, payments engines, customer platforms and specialist applications have been implemented, customised and integrated over many years. Mergers, acquisitions and regulatory change have only added to the complexity.
BFSI organisations rarely suffer from a shortage of information. They hold vast quantities of customer, transaction, product, risk and operational data. The challenge is that much of this information remains fragmented across disconnected systems and business domains.
Humans have learned to work around these limitations.
AI agents cannot. A pilot can succeed using a carefully curated subset of data. Scaling AI across customer servicing, lending, payments, fraud detection, compliance and product management requires something fundamentally different. It requires information that is accessible, connected and available when and where AI systems need it.
For many organisations, that remains a significant challenge.
Trustworthy data is necessary. Accessible data is essential.
Financial institutions operate within one of the most heavily regulated environments in the economy. They must be able to demonstrate where information originated, how it as transformed and whether it can be trusted.
We explore those governance, lineage and provenance challenges in more detail in our companion article, "Can your organisation prove where its data came from?"
Those capabilities are critical. But governance alone does not solve the AI challenge.
An organisation may have excellent visibility into its data lineage and still struggle to deliver meaningful AI outcomes if critical information remains trapped inside disconnected systems, ageing integration platforms and application silos.
AI needs trustworthy data. It also needs accessible data. The ability to understand the journey of information is important. The ability to move that information securely and efficiently across the organisation is equally important.
Why AI raises the stakes
Historically, fragmented data and brittle integrations created inefficiency. An experienced lending specialist could reconcile information across multiple systems. A fraud analyst could investigate anomalies. Operations teams could manually resolve discrepancies and manage exceptions.
These limitations were frustrating, but they were survivable because people compensated for them. Agentic AI changes the equation. When AI-powered processes are making recommendations, orchestrating workflows or initiating actions across multiple systems, every disconnect in the underlying architecture becomes a scaling constraint. The issue is no longer whether the data exists. The issue is whether the right information can be discovered, retrieved and acted upon quickly enough.
An AI-powered customer onboarding journey is only as effective as the systems it can access. A fraud detection agent is only as effective as the information it can correlate.
An underwriting assistant is only as effective as the data it can see. The bottleneck is increasingly becoming integration rather than intelligence.
AI is moving faster than enterprise architecture
Much of the AI conversation still focuses on models. That's understandable. New capabilities appear almost weekly, and every new release seems to reset expectations about what AI can achieve.
But while organisations are debating the capabilities of the latest models, a quieter shift is already underway. Agents are moving into production. Organisations are increasingly deploying AI systems that can retrieve information, interact with enterprise applications, coordinate work and initiate actions on behalf of users. Standards such as Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication are accelerating this shift by making it easier for AI systems to connect with enterprise data, services, business processes and each other.
Together, these emerging standards are helping organisations move from isolated AI assistants towards coordinated ecosystems of specialised agents that can discover capabilities, exchange context and collaborate to achieve business outcomes.
For BFSI organisations, this creates enormous opportunities. Imagine onboarding agents coordinating with compliance agents to accelerate customer acquisition. Fraud detection agents collaborating with payments and transaction monitoring systems to identify suspicious behaviour more effectively. Lending assistants gathering information from multiple business systems and presenting recommendations to specialists, allowing them to focus on judgement rather than administration.
These scenarios are no longer theoretical. But they place significantly greater demands on the underlying technology landscape. Every new AI capability depends on the organisation's ability to securely access information, integrate systems and orchestrate business processes. As MCP and A2A adoption grows, organisations need to think beyond how a single AI agent interacts with a single system. They need to consider how entire networks of agents will access, exchange and act upon information across the enterprise.
Every interaction between an agent and a business system crosses application, data and security boundaries. Every interaction between agents introduces new dependencies on interoperability, discoverability and governance. Every action must operate within established controls and risk management frameworks.
What begins as an AI initiative quickly becomes a data and integration challenge. The question for BFSI organisations is no longer whether AI agents can deliver value. The question is whether the underlying architecture is prepared to support them.
- Can critical information be accessed when it is needed?
- Can systems interact without fragile point-to-point integrations?
- Can new AI-enabled capabilities be introduced without creating additional complexity and risk?
As AI moves from answering questions to performing work, the answers to those questions become increasingly important.
The ROI trap
One of the reasons these foundational challenges remain unresolved is that they are difficult to justify through the business case of any individual project.
A customer onboarding initiative may benefit from improved integration. A fraud modernisation programme may benefit from better information accessibility. A lending transformation may benefit from both. Yet each initiative is typically expected to justify foundational investment independently. That's the trap.
The ROI of modernising data and integration foundations rarely sits within a single programme of work. It emerges across an entire portfolio of transformation initiatives.
The value is not found in one project delivering slightly faster. The value is found in every future project delivering faster. When organisations establish connected data ecosystems, interoperable platforms and modern integration architectures, the benefits compound:
- AI use cases move from concept to production more quickly.
- Customer onboarding becomes more efficient.
- Fraud and AML capabilities become more effective.
- Open Banking initiatives become easier to support.
- Product innovation accelerates.
- Regulatory change becomes easier to implement.
- Agentic AI initiatives become practical rather than aspirational.
The stretch goals stop being stretch goals. They become quick wins. Innovation at scale. Regulatory compliance at scale. Customer data quality at scale. Transformation at scale.
What AI-ready foundations actually require
Organisations successfully scaling AI are not treating data and integration as technical housekeeping activities that can be deferred until later. They are treating them as strategic enablers of their AI agenda.
In practice, this means:
- Building integration architectures that provide secure, real-time access to business information.
- Reducing dependency on batch-based data movement wherever appropriate.
- Establishing interoperable platforms that enable information sharing across business domains.
- Creating secure and governed mechanisms for agent interactions.
- Modernising legacy integration approaches that impede innovation and increase delivery complexity.
- Developing the specialist skills and delivery capacity required to support AI initiatives at enterprise scale.
These capabilities do not slow innovation down. They allow organisations to innovate faster without continuously rebuilding the same foundations.
The question BFSI leaders should be asking
The question is no longer whether AI belongs in the enterprise. The question is whether the enterprise is ready for AI. Can your organisation make the right information available to AI systems when it is needed? Can your existing integration landscape support a future built around intelligent automation and agent orchestration? Can your current architecture enable innovation at the pace your business expects?
If the answer to any of those questions is uncertain, the risk is not that AI will fail.
The risk is that AI succeeds faster than your technology foundations can support it. Fusion5 helps banks, insurers and financial services organisations strengthen the data, integration and interoperability capabilities required to support AI at scale. By modernising the foundations beneath AI initiatives, organisations can unlock faster delivery, stronger operational resilience and greater confidence in the opportunities that lie ahead.
If you're assessing your organisation's readiness for the next wave of AI adoption, now is the time to evaluate whether your data and integration foundations are ready to support it.
Sean Mosely | COE Lead - Architecture - Data & Integration NZ
Sean specialises in connecting strategy, architecture, and technology to help organisations simplify complexity and deliver lasting business outcomes. He brings a pragmatic perspective on integration, modernisation, and AI adoption, grounded in real-world experience.
