AI has made finance transformation surprisingly easy to demonstrate. Connect a model to financial data, ask a question in natural language, generate a variance explanation, produce a forecast or draft a management report, and within a few minutes it can look as if the finance function has been reinvented.
The reality is more complicated.
What we see today, copilots, agents, automated analysis, conversational interfaces and intelligent reporting, is only the visible part of a much larger transformation. I tend to think about it as an iceberg. The applications are what everyone sees above the waterline, while underneath sit the things finance organizations have been wrestling with for decades: processes, data definitions, systems, governance, controls, ownership, skills and change management.
AI does not reduce the importance of those foundations. If anything, it makes them more important.
There is now enough evidence to move the conversation beyond theoretical use cases. McKinsey has described a global consumer-goods company using generative AI to help finance teams explain budget variances, with the company estimating that the technology saves finance professionals roughly 30% of their time. In another example, a biotech company used AI to continuously compare contracts and invoices and identified contract leakage equivalent to around 4% of spend. These are not simply chatbot experiments. They affect the economics of the business.
Treasury is moving in a similar direction, with AI increasingly applied to cash forecasting, liquidity management, foreign-exchange exposure and fraud detection. Planning teams are beginning to combine financial, commercial and operational data to understand performance drivers and generate scenarios, rather than simply producing another forecast.
The direction is becoming clearer. Finance is starting to move from automating isolated tasks toward systems that can investigate, reason, recommend and, in some cases, execute parts of a workflow. That is a meaningful shift, but it also exposes every weakness in the underlying finance environment.
Consider something as ordinary as contribution margin. Ask five businesses how they calculate it and you may receive five different answers. In a large industrial company, you may even receive different answers across business units. Which freight costs are included? Are rebates deducted? How is inventory absorption treated? Which allocation rules apply? How are intercompany transactions handled?
Experienced FP&A teams learn these rules over time. Much of that knowledge is rarely captured in one place. It lives in spreadsheets, reporting routines, ERP configurations, management habits and the heads of people who have worked in the company for years. An AI system needs access to the same context. Otherwise, it can produce an eloquent answer that is technically plausible but economically wrong.
The same challenge appears in industrial finance. An agent may be perfectly capable of explaining purchase-price variance, but the explanation is only useful if the system understands the material hierarchy, standard-cost methodology, plant structure, currency effects and the way the company has configured its ERP. Finance is full of these details. They are often invisible until they are wrong.
This is one of the reasons I believe the foundations matter so much. The language model may be probabilistic, but the ledger cannot be.
A serious AI-enabled finance transformation therefore starts below the surface. It begins with process clarity. If a monthly performance review requires fifteen people to spend days reconciling numbers across systems before anyone can discuss what actually happened in the business, putting AI on top of that process will not automatically create a high-performing finance function. The underlying workflow needs to be reconsidered.
The same is true for data. Finance needs more than access to tables. It needs consistent definitions, hierarchies, mappings, business rules and relationships between financial and operational information. Revenue needs to mean the same thing across systems. Customers, products, plants, cost centers and legal entities need to connect in a way that reflects how the business is actually managed. Metrics need clear definitions and ownership.
Governance becomes equally important once AI starts influencing decisions. If an AI-generated analysis contributes to a pricing decision, a forecast, a capital-allocation discussion or an external report, finance needs to understand how that conclusion was produced. Source lineage, permissions, reconciliations, approval rules and auditability cannot be treated as features to add later.
There is also a human side to this transformation that is easy to underestimate. If AI can investigate data, prepare analysis and generate explanations much faster than today, the role of the finance professional inevitably changes. Organizations will need to decide who reviews outputs, what an agent is allowed to do autonomously, where human approval is mandatory, which activities should disappear entirely and which skills become more valuable.
These are not simply technology decisions. They are operating-model decisions.
Recent research points in the same direction. KPMG’s 2026 Global AI in Finance study found that active use of AI across finance has expanded quickly, but only a minority of organizations report that it is significantly exceeding expectations. Data quality, integration and interoperability remain major constraints, while organizations with stronger governance and better audit evidence appear to achieve better outcomes.
That does not mean finance leaders should wait until every process and every dataset is perfect. In fact, I think that would be another mistake. Many organizations have spent years discussing harmonisation, architecture and transformation without delivering enough tangible value to the business.
A better approach is often to start with a real finance problem and build the foundation around it. Take margin analysis, working capital, forecasting or the monthly operating review. Connect the data required for that problem, define the business logic properly, establish the right controls, redesign the workflow and put the capability into the hands of people who actually make decisions.
Then expand.
Done well, this creates something much more valuable than a collection of disconnected AI pilots. Each use case improves the underlying financial context, data model and governance framework, making the next use case easier and more powerful. Over time, finance begins to develop something closer to an intelligent operating system for understanding the business.
For years, finance transformation programs have promised that finance professionals would spend less time gathering information and more time helping the organization decide what to do. The promise has been repeated so often that it has almost become a cliché.
AI may finally make that possible at a very different scale.
However, the ambition should not simply be an autonomous finance department where agents perform today’s processes faster. That would be a surprisingly small outcome given the potential of the technology. The bigger opportunity is to rethink how finance understands the business: continuously connecting financial and operational signals, investigating what is changing, explaining why it is changing, modelling what could happen next and helping management decide what to do about it.
That requires powerful AI, but it also requires very good finance.
The companies that understand both sides of the iceberg will be the ones that create lasting value from it.
