Finance needs to transform before AI enters the picture. Here’s why.

The teams that win the AI race in finance will be the ones that build their infrastructure now, not the ones that wait for the models to be ready.

AI comes up in almost every conversation I have with finance leaders right now. Interestingly, it seems that less than 10% of CFOs have fully integrated or scaled AI use cases across their organisations.¹

What strikes me is how differently people are responding to it. Some are already running pilots and looking into rebuilding their data architecture. Others are watching carefully, waiting for ‌models to mature before they commit. I can see the logic in both positions, but the outcomes they’re producing are starting to diverge.

The organisations holding back tend to share a common profile. They're tied to vendor contracts signed years ago, operating inside hierarchies where finance leaders often feel they lack the mandate to affect change, and running on fragmented systems that’d take significant effort to rebuild. These are successful businesses that have accumulated these constraints while growing. But now those constraints are compounding into something harder to shift each year.

I understand the instinct to wait. Our function exists to produce accurate numbers that people trust and use to make decisions. That’s why the idea of introducing a tool that may make mistakes feels like too big a risk. But waiting has an opportunity cost that won’t show up on any dashboard until it's too late to close the gap.

Waiting for perfection is a trap

Finance has to be accurate, and AI isn't there yet. As finance leaders, our accuracy instinct pushes us to hold off until the outputs are trustworthy. But AI doesn't improve the way most people assume it will. There's no gradual, linear climb toward reliability that you can track and time your entry around.

What I've observed is something closer to a long plateau of imperfect outputs, followed by a fairly sudden jump in capability. I think of this as the bad-bad-not-bad curve.

AI will never arrive perfect out of the box. It needs to be trained. For AI to reach an appropriately accurate level on its own, it needs a period of continuous human oversight and fact-checking. When you give AI feedback and input, it learns your business and learns how to create the right output for you. By experimenting with AI, even while the outputs are imperfect, you’ll get a head start in reaching that jump in capability.

The fundamental divide happening in the industry

Here’s what most finance teams look like today: transactional data sitting in one tool, expense data in another, FX in a third, with a spreadsheet reconciling all of it. A Power BI report built 18 months ago that nobody wants to touch now. A finance leader who knows exactly what needs to change, but is waiting for a quarter where there's enough bandwidth to start. Meanwhile, the data compounds in complexity, the workarounds multiply, and the cost of rebuilding climbs a little higher every month. Traditional organisations are weighed down by the legacy debt in their software and mindset.

Teams that are more agile occupy a different position. The organisational layers aren't so deep that a motivated finance leader can't effect change. The vendor relationships aren't so politically loaded that consolidation requires a board-level mandate. There's still real room to move. But that room shrinks every year as the stack grows, the team builds new habits around existing tools, and the business adds entities, currencies, and markets that each bring their own data trail.

The organisations moving fastest on AI right now are the ones who’ve recognised this window and started consolidating before the urgency felt obvious. When AI enters the picture, it widens the gap between those leading and those falling behind.

The right infrastructure amplifies AI’s value

In my experience, there are three parts of your infrastructure that affect your success in implementing AI:

1. Data that the AI model can trust

AI trusts the data it sees. It produces outputs with the same confidence regardless of whether the underlying data is clean. If you have inconsistent transaction records, a chart of accounts without a governing logic, or unreconciled multi-currency data, AI will give you answers that look authoritative but may be wrong. Most teams only discover the problem after they’ve already made a mistake. Early adopters treat data hygiene as a foundation. They clean, unify, and standardise before scaling AI, giving them a head start.

2. Systems that share a common view

Finance teams often work across several disconnected tools: payables, expenses, FX, and reporting. When data lives across platforms that don't communicate, AI can only generate insights based on fragments of the business, not the whole picture. Teams that connect their data create a single source of truth that AI needs to run autonomously.

3. Processes that are designed for automation

Manual, ad-hoc workflows can’t simply be handed off to AI. If your month-end close involves someone downloading a CSV, reconciling it against another file, and routing the result through an email approval chain, introducing AI into that workflow surfaces the complexity without resolving it. Automation will amplify the architecture that you already have. If your foundations are strong, it scales. If they’re not, it cracks. To get real value from AI, the process architecture has to change first, with processes that are defined, repeatable, and structured for AI from the start.

Getting these three things right doesn't require perfect infrastructure before you begin. What it does require is an honest read of your current infrastructure, because the answer shapes what kind of finance leader you need to become.

AI-readiness starts from finance leaders

The skills that made a finance leader effective five years ago are shifting in ways that most people haven't fully reckoned with yet.

Finance has always been analytical. We've been trained to work with data, to find the signal, to produce the numbers the business needs to make a decision. That capability remains valuable, but the nature of the work is changing. The finance leaders getting ahead now are building and refining the systems that perform the analysis, rather than performing it themselves. Instead of running a campaign's ROI analysis, you build and train the AI agent that runs it, and then your job becomes improving the agent.

Forward-thinking finance teams are already letting AI agents run their finance operations. Think touchless, autonomous financial operations. Documents are ingested, categorised, and reconciled by agents working in the background. Approvals route themselves. Exceptions surface for human review. The human role in the workflow is setting the parameters and refining the logic those agents run on.

The metric that matters is time to insight. Reporting accuracy at month-end is important, so is how quickly the business can detect a shift, understand what's driving it, and respond. AI can radically reduce decision latency, the gap between something appearing in your data and your team acting on it. Closing that gap requires connected data and orchestrated systems that create a foundation that AI can learn from.

Build now, not when it's perfect

Agility is becoming as important as accuracy. We need to recognise that the ability to move, learn, and adapt systems is now as important as the precision of any individual output. The teams waiting for AI to reach some threshold of reliability before they start using it are making the same mistake as ‌organisations that delayed their digital transformation because the cloud felt unproven. They were right to notice the imperfection. They were wrong about the cost of waiting.

What you need today is a foundation that'll let you use AI well when it matters. Clean data, unified systems, and digitised processes are finance fundamentals. The benefits of getting them right and the costs of getting them wrong are growing faster than most finance leaders expect.

We’ve spent the last decade building a unified financial infrastructure, one place where payments, FX, spend, and data all live together, to give finance teams a foundation they could build on. That same consolidation is what makes the move toward autonomous finance a practical next step rather than a distant ambition.

The models will catch up. The question is whether your infrastructure will be ready when they do.

 ¹https://www.egonzehnder.com/press-release/cfos-recognize-ai-s-strategic-importance-but-adoption-remains-in-early-stages-new-egon-zehnder-survey-shows