The conversation around AI in finance has changed dramatically over the past year. Not long ago, nearly every executive meeting revolved around a single question:

"What's our AI strategy?"

Today, the conversation is far more practical:

"We are spending millions on AI. Where's the return?"

That's a healthier discussion.

AI has moved well beyond experimentation. Finance organisations are investing heavily in Microsoft Copilot, enterprise language models, AI capabilities embedded within ERP and planning platforms, implementation partners, cloud infrastructure, governance, and internal AI teams. The investment is substantial, and CFOs are right to expect measurable returns. Yet despite increasingly capable models, many organisations aren't making dramatically better decisions.

I don't think that's because AI isn't intelligent enough. I think we're blaming intelligence for what is fundamentally an Enterprise Context problem. Much of the discussion surrounding enterprise AI assumes the next leap in value will come from more capable models, perhaps even Artificial General Intelligence (AGI). Better reasoning. Better planning. Better coding. Better analysis. I'm not convinced that's where the biggest gains in corporate finance will come from.

Most finance work isn't constrained by a lack of intelligence. Invoice matching, bank reconciliations, purchase order validation, revenue variance analysis, cash forecasting, and close activities are highly structured processes. They depend far more on consistent data, standardised business rules, reliable master data, and well-governed processes than they do on extraordinary reasoning. If those foundations are weak, even the most capable AI model has very little to work with. For many finance workflows, the real constraint isn't model capability. It's the quality of the Enterprise Context surrounding the model.

Here's a thought experiment that makes the distinction concrete. Imagine hiring the best FP&A leader in the world. On their first day, you hand them five different revenue numbers, duplicate customer records, inconsistent product hierarchies, a broken chart of accounts, unreconciled spreadsheets, and missing invoices.

You haven't hired the wrong person. The problem isn't their analytical ability. Before they can improve forecasting, recommend capital allocation decisions, or identify operational opportunities, they first have to reconstruct the organisation's version of reality. They need to determine which numbers can be trusted, which reports are authoritative, and which processes actually reflect how the business operates.

Now replace that finance leader with the world's most advanced AI model. The conclusion barely changes. No amount of intelligence, whether human or artificial, consistently overcomes unreliable information. Intelligence is only useful after the organisation has established sufficient trusted context for it to reason effectively. Without it, humans and AI encounter the same obstacle. AI simply reaches it much faster.

That thought experiment also explains why I think many organisations misunderstand what AI is actually revealing. For years, finance teams quietly built workarounds that kept the business moving. Someone knew which ERP report couldn't be trusted. Someone else maintained the spreadsheet that corrected inventory allocations before every month-end close. Treasury manually adjusted cash balances every Friday because one application consistently lagged behind another. FP&A carried out offline adjustments that everyone accepted, but no one formally documented.

None of those workarounds appeared in an organisation chart, a process map, or an ERP implementation guide, yet they became part of how the business actually operated. Experienced finance professionals learned which reports to trust, which business units used slightly different definitions, which adjustments always appeared during close, and which exceptions never made it into formal documentation. Over time, those informal practices became part of the organisation's Enterprise Context: the collective knowledge that allowed people to navigate imperfect systems without consciously thinking about it.

AI doesn't inherit that knowledge. It doesn't know which report everyone quietly ignores, the spreadsheet hidden on someone's desktop, or the unwritten rules experienced employees have accumulated over the years. When AI enters the organisation, those hidden dependencies become visible almost immediately. What initially looks like an AI implementation problem is often years of accumulated organisational and process debt finally being exposed. AI isn't breaking finance. It's revealing where finance was already fragile.

Looking at AI through this lens also changes how I think about value creation. Most conversations reduce AI success to model capability, but I increasingly think of it as a multiplication problem rather than a technology problem.

Business Value = AI Capability × Data Quality × Process Maturity × Organisational Adoption

The multiplication is important because weakness in any one dimension limits the value created by all the others. A world-class model cannot compensate for poor master data. Reliable data won't generate meaningful value if business processes remain inconsistent, and well-designed processes won't produce impact if finance teams don't trust or adopt the technology. Model capability certainly matters, but it is only one multiplier within a much larger system.

Framed this way, Enterprise Context isn't simply another input into AI. It's the environment that determines whether intelligence can produce value at all. Organisations that invest exclusively in better models while neglecting their data, governance, and operating processes shouldn't be surprised when AI produces disappointing returns. The technology is functioning exactly as designed; it simply doesn't have enough trusted context to make consistently good decisions.

Five Questions to Ask Before Expecting AI ROI

Before investing further or scaling an AI use case, FP&A leaders should assess whether the necessary Enterprise Context is already in place.

Table 1. Five Questions to Ask Before Expecting AI ROI

Weakness in any one of these areas can place a ceiling on the value AI can generate.

If Enterprise Context is the constraint today, what happens once organisations solve it? The opportunity becomes much larger than automation. For decades, finance has evolved through three broad stages. First, we recorded transactions. Then we reported performance. More recently, we've become increasingly focused on predicting future outcomes. I believe the next evolution isn't prediction. It's simulation.

Simulation isn't simply a better forecasting technique. It's a fundamentally different way of supporting decisions. Instead of asking, "What happened?" or "What will probably happen?" finance increasingly asks a different question:

"What happens if we make this decision?"

That capability only becomes possible once organisations have built a strong Enterprise Context. A Finance Digital Twin built on five conflicting revenue numbers and a broken chart of accounts is simply a faster way to simulate bad data. The Digital Twin isn't a substitute for fixing the foundation; it's what becomes possible once the foundation is fixed.

For the past two years, the conversation has largely centred on one:

"How do we get AI into finance?"

I'm beginning to think that's the wrong question.

A better one is:

"Is our finance organisation ready for AI?"

Those are fundamentally different conversations. The first assumes technology is the constraint. The second recognises that organisational readiness is. AI models will continue becoming more capable. But unless finance organisations improve the quality of their data, standardise their processes, preserve institutional knowledge, strengthen governance, and connect business context across the enterprise, better models alone won't consistently produce better decisions.

The organisations that create the greatest value from AI won't necessarily have the smartest models. They'll be the organisations that finally give their best FP&A leader, human or artificial, something they've rarely had: a chart of accounts, a customer master, and a set of numbers worth trusting. For decades, finance competed on intelligence. In the AI era, it will compete on Enterprise Context.