Why AI is still delivering in parts of finance, not across it

Spend time with finance leaders at the moment and the conversation tends to circle the same set of questions, usually centred on which AI tools are worth adopting, what other organisations are doing, and where the real value is starting to show up. The market is not short on options, every platform has introduced some form of AI, new tools appear constantly, and most finance teams have already started experimenting in small but meaningful ways. 

Commentary is being drafted faster, variances are explained more easily, forecasts can be generated and adjusted in a fraction of the time, and routine tasks that once took hours now take minutes, particularly when the starting point is already structured data that the system can interpret reliably. 

In many cases, that initial experience is positive enough to justify further exploration, which is why AI has quickly moved from something finance teams were observing to something they are actively using across different parts of the function. 

The limitation tends to appear when AI moves beyond isolated tasks and into the core of how finance actually operates, where expectations are higher, dependencies are broader, and the margin for error becomes significantly lower than it is in controlled, one-off use cases. 

Take month-end reporting as a simple example, an area where many teams have already started to experiment with AI. Generating commentary is now straightforward, movements across accounts can be explained quickly, trends can be identified, and the output often reads as though it has been carefully prepared by an experienced analyst. The moment that output is relied on, a different set of questions begins to surface, not about how it reads, but about where it comes from, which system it reflects, and whether it aligns with the numbers the business is using elsewhere. 

In many organisations, those questions are not easy to answer because the data itself is not unified, with finance pulling from the general ledger, operations holding inventory and fulfilment data, sales activity sitting in CRM, and adjustments often happening outside formal systems, usually in spreadsheets that introduce slight variations each time they are used. 

AI reflects that environment, working with whatever it is given, which means inconsistencies in the inputs inevitably carry through into the outputs. 

This becomes even more visible in processes such as reconciliation, where AI performs well when data is structured and aligned, matching transactions quickly, identifying patterns and surfacing exceptions with minimal effort. The same capability becomes far less reliable when transactions arrive from multiple sources with different formats, inconsistent timing and incomplete references, at which point the work shifts back to the finance team to resolve what the system cannot confidently align. 

A similar pattern emerges in forecasting, where AI-driven models can adjust projections based on new data and generate scenarios quickly, providing a strong starting point for analysis. The speed is impressive, though the usefulness depends heavily on the consistency of the underlying data and assumptions, and where those vary across business units or historical data is incomplete, finance teams still need to apply judgement, refine outputs and validate results before anything can be relied on. 

At this point, the conversation begins to shift away from the capabilities of AI itself and towards the environment it is operating within, where the real constraint starts to become visible. The challenge is no longer access to AI tools, it is whether the surrounding data and processes are structured in a way that allows those tools to deliver outcomes that are both meaningful and reliable.  

Data consistency becomes a limiting factor, as does process clarity, because AI performs best when it can follow a defined flow, understand what triggers an action, apply rules consistently and support the next step in a process. Many finance workflows are well understood in practice, though they are often shaped by people rather than structured in a way that a system can follow, with a significant portion of the logic sitting in individual knowledge rather than in defined rules. 

Governance adds another layer to this, as finance operates on traceability, where every number must have a source, every adjustment must be explainable and every decision must be accountable. Introducing AI into that environment raises practical questions around how outputs are generated, how actions are triggered and how both can be audited, and where those answers are not clear, adoption tends to slow regardless of how capable the technology appears. 

This is why many finance teams feel as though they are using AI without seeing a meaningful shift in how they operate, as the tools themselves are working and the environment around them is not yet structured to support broader change. 

The organisations starting to move beyond this are approaching the problem from a different angle, focusing less on the tools themselves and more on the structure that surrounds them. Data is being consolidated into systems where finance, operations and transactional activity sit together, definitions are aligned, timing becomes consistent, and the effort required to reconcile across systems begins to reduce. 

Processes are becoming more explicit, approval paths are clearly defined, decision points are understood, and the flow of work becomes something the system can support consistently rather than something that relies on individual interpretation. Within that environment, AI begins to behave differently, working with live data, operating within defined workflows and supporting processes without stepping outside the controls that finance teams depend on. 

The shift is less about introducing something new and more about enabling the system to do more of the work it already supports, which is where the real impact begins to show. 

For finance leaders, this reframes the question in a practical way, moving the focus away from which AI tools to adopt and towards whether the organisation is ready for AI to operate within its systems and processes. That readiness shows up in the quality and consistency of data, the clarity of workflows and the strength of governance around decision-making. 

Teams that address these areas tend to see AI deliver quickly and predictably, while others remain in a cycle of experimentation, finding pockets of value without a broader shift in how finance operates. 

In an upcoming session with Oracle NetSuite, AI for finance leaders - what to use now and what's next, taking place on 1 April 2026 at 12:00 pm AEDT, we will explore how finance teams are addressing these challenges in practice, where AI is already delivering value inside core workflows and what needs to be in place for it to work effectively. 

Register here to join the session. 

If you are seeing AI work in parts of your finance function and fall short in others, this is a useful place to start.