Somewhere in the world, there’s a finance team that’s a little ahead of you. They don’t necessarily have a data engineer or an AI strategist, and their stack looks basically the same as yours.
They’re ahead because their head of FP&A spent an afternoon trying something new with AI. They asked a lot of questions of a tool they didn’t fully understand yet. What they got back wasn’t perfect, but it was useful enough to share with the team. Then the team started trying it too. Now they’re using AI to automate work their competitors are still wasting hours on, while spending more time on strategy.
Their real advantage was a single afternoon. Not a platform migration or a six-month roadmap; just a few hours of experimentation that changed how they work (slightly) from that point on.
Right now, the gap between teams that are getting actual value from AI and the teams that aren’t is small. Much smaller than most people assume actually. But it’s widening. The good news is that it has almost nothing to do with technology, and you can be on the right side soon.
The gap most teams don’t bridge
When most people talk about AI in finance, they start with infrastructure. They talk about cleaner data, better pipelines, the right platform and maybe even a data engineer. All of those things are useful, but they often come with an assumption: You might not be ready yet because you haven’t built the right foundation.
Foundations matter, but teams getting real value from AI might not be doing everything right on paper either. What they’re doing differently is trying things before they fully understand them. So to begin, all you have to do is sit with a tool that doesn’t work perfectly yet, and keep going.
That can be uncomfortable, because finance is built on a different instinct. You have to be precise and defensible. You never share a number you can’t stand behind, and you definitely don’t send a model to the board that you haven’t stress-tested six ways. If you’ve worked in finance, you know this in your bones: you never present something you’re not 100 percent sure about. That instinct is part of what makes finance reliable, but it also slows AI adoption.
The real constraint, then, is whether your team can tolerate some uncertainty while learning. Working with AI doesn’t feel like learning a new system. There’s no clean handoff from “trained” to “operating.” You try something, you see where it helps, you notice where it breaks and you adjust. Over time, you develop a feel for when it needs more context and when it can surprise you. From the outside, that process can look messy and hard to explain in a status update, which is exactly why it’s easy to avoid, and why it ends up mattering so much.
What AI adoption looks like in practice
That head of FP&A sat down on an ordinary afternoon and opened an AI tool. Maybe it’s something the company already pays for or something they’ve been meaning to try.
They ask a simple question: Why did marketing spend come in 12 percent over plan last quarter? The first answer doesn’t seem quite right. The direction is there, but the framing feels off, so their instinct is to close the tab. This isn’t reliable, they think, I can’t use this.
But they persist: ask again, add more context this time, narrow the scope a little and ask AI to focus on the largest drivers. Now the answer is better. Still not something they’d send as-is, but there’s a noticeable improvement. They spend a few minutes shaping it and adjusting the prompt, and end up with a draft they can use. Something that would’ve taken way longer to write from scratch.
That specific moment when you get a rough first output and decide to stay with it, instead of moving on, is what matters with AI. So why do most teams never get there? Because they try once, see something imperfect and take it as evidence that the tool isn’t ready yet. Another team tries the same thing, gets the same imperfect-looking answer, but spends some time trying to fix it.
Over time, those small differences compound and show up in how work happens. Some orgs have access to every new tool. They’ve run workshops and written memos and even assigned someone to be responsible for AI, but the work itself looks the same. Forecast reviews run the same way as before, and variance explanations still get assembled at the last minute. When the tools change but the habits don’t, not much else moves either.
Why this is harder than it sounds
I think the “just be curious” framing can sound glib if you don’t acknowledge what’s underneath it. Finance leaders don’t experiment because being wrong carries a huge cost. It’s both financially and politically expensive.
If you try an AI-generated analysis and a bad number makes it into a board deck, it doesn’t matter how it got there. You’re the finance person, so you’re the one who owns it. Nobody is going to say, “Well, she was experimenting.” They’re only going to say the numbers are wrong. And over time, you internalize this. You build a career on being the person in the room whose numbers can be trusted. Naturally, experimentation feels risky because it threatens the very thing that makes you valuable. (This is a much deeper problem than cultural change. It’s about identity, and the fact that finance is one of the few functions where a single wrong number can actually haunt you for years.)
Does this mean early adopters are reckless or not concerned with accuracy? I don’t think so. They’ve just found ways to be uncertain without taking on the full risk. They try AI on last quarter’s closed data, where nothing can move, before running it on this quarter’s live data. They let it generate a first pass on a variance explanation for a low-stakes meeting, where the output will be checked by hand anyway.
It’s a way of creating space to learn, which only comes from experience. So the early work is figuring out where you can afford to be wrong, so you can start learning what it looks like when it’s right. Nothing about that is reckless.
You could automate actual work today
If that feels theoretical, look at the actual work that gets done. Think about your last variance analysis: You pulled data from your CRM or ERP, exported it to a spreadsheet, built a pivot, cross-checked line items and looked for patterns. Then you wrote up what happened and why. That last part usually takes the longest. It’s easy to lose an entire afternoon to it.
Now imagine starting in a different place, where you ask: Why did marketing spend come in over plan last quarter? The system surfaces the main drivers and drafts a clean explanation. Or: What do pipeline trends look like over the last 90 days? It writes the query, pulls from the CRM and gives you something to work with.
That sounds easy, and it actually is. A lot of teams could do this today with what they already have. And it doesn’t reduce the role of the finance leader. If anything, it makes them more valuable. It’s just the manual work and the first pass at an explanation that get compressed. What remains is deciding what matters, and how to frame it for the CFO or the board. That part still takes time and still depends on judgment. The difference is you spend more of your day there, and less of it rebuilding the same analysis you ran last month.
The tool is becoming free. The judgment is still scarce. And the constraint is smaller than it looks; it’s mostly just the decision to try. No SQL or technical skills required.
Curiosity will take you far
That head of FP&A, she’s not more technical than her peers. And she didn’t do anything dramatic, like getting budget approval or waiting for engineering to greenlight a platform. She just spent an afternoon playing with a tool.
What that afternoon actually bought was a feel for where the model helps and where it struggles. She now knows which questions to ask and what to trust vs. verify. Over time, that turns into something that looks like intuition. Now her team has started building it too.
That intuition comes from curiosity. From the habit of asking: Can this thing do part of my work? What happens if I give it better context? What breaks if I push it further? Where is it surprisingly good, and where is it obviously dangerous?
That knowledge compounds. A year or two from now, they’ll have spent that time building a feel for what’s worth doing and what isn’t. They’ll have thousands of experiments, and a clear sense for where AI actually accelerates their work. Meanwhile, the team waiting for the right level of data readiness or the right tool will be two years behind, starting from zero.
That’s the gap: actual, real-world experience. And it starts small, with an ordinary afternoon where one person is curious enough to push past uncertainty.
