Many conversations I have with finance leaders these days inevitably revolve around which AI tool they're piloting next, how their team adopted it, and how fast it's running. And while we all feel a general sense of progress, discussions about the true ROI of AI in finance feel vague.
I've spent years measuring technology ROI in finance, through ERP rollouts, automation waves, and data warehouse rebuilds. I think the problem is that we're measuring AI the same way by default: faster cycle times, lower transaction costs, or headcount avoided.
Automation made existing work faster. And while AI can deliver that value too, often to a much greater degree, it also can completely change what work gets done and what decisions get made because of it.
So, if we want to understand and measure the true ROI of AI in the finance function, we need to start explicitly measuring the strategic impact of these tools, not just their efficiency and productivity benefits.
Table Stakes ROI Metrics Aren’t Enough to Measure the Value of AI
Calling efficiency and productivity benefits "table stakes" isn't meant to devalue them. They're real, measurable forms of ROI that finance teams can already track and understand. But if we think of them as the finish line, we put ourselves in a position to gloss over so much value that AI delivers.
I like to think about AI ROI for finance teams in terms of a three-layer framework to get a complete picture. Each of these layers builds on the one before it.
Layer 1: Efficiency Metrics. Doing the same work faster. Close cycle time dropping from 10 days to five, cost per transaction falling on AP and payroll.
Layer 2: Productivity Metrics. Doing more work with the same team. Revenue per finance employee climbing, one analyst handling three business units instead of one.
Layer 3: Strategic Impact Metrics. Using AI to change what decisions get made and how fast. This is the layer almost nobody measures reliably, and it's the one that actually moves the business.
For accounting teams, Layers 1 and 2 are probably enough. The job is accuracy and control, not invention. If AI gets my close down from 10 days to five, or flags a reconciliation error before it compounds into something bigger, it's doing exactly what I need from accounting. I want that team to be predictable, not necessarily creative. The moment my accountants start getting creative, I have a different problem on my hands.
FP&A is where I want creativity, and that's exactly where Layers 1 and 2 stop being enough.
The job is to see what's coming and get in front of it, which means an FP&A team that can only point to faster models and cleaner data has automated the function without actually changing it. I've watched that happen, where a team gets faster at producing the same reports nobody acts on differently. Layer 3 is where AI starts pulling its real weight, and it's where I want to take you next.
What It Looks Like to Track the Strategic Impact of AI
Efficiency and productivity metrics are easy to track because they look roughly the same for every finance org — close cycle time is close cycle time wherever you work. Strategic impact doesn't work that way, though.
It takes a different shape at every company, built out of whatever decision or opportunity was specific to that business at that moment. That’s why a decision getting made three weeks faster or an AI finance tool helping you spot an opening in a new market won't translate into a clean ROI figure you can drop into a board deck. You might wait a year or more before either one shows up in the numbers at all.
But none of this is a reason to give up on measuring the strategic impact of an AI use case for finance. It just means we have to match what happened to the value it created instead of stopping short at efficiency or productivity metrics.
I have two examples from my own experience that show what that looks like in practice.
Example 1: One FP&A Hire for Five Business Partners
As a finance org scales, headcount usually scales with it. More business partners typically means more FP&A analysts, often close to one per relationship. At a previous company, I tested whether that ratio still held once AI was in the mix. I hired one analyst, gave her AI tools, and had her support five VPs instead of one.
Two things had to be true for that to work. She needed real-time access to data instead of waiting on the monthly close, and she needed to be able to model a decision on the spot instead of promising a follow-up.
She pulled live data directly from accounting rather than waiting on close
She could model a headcount swap or budget shift in the meeting itself
She supported five VPs in roughly the time the traditional model would have given to one
Under the old model, a VP wanting to swap a budgeted headcount for a stronger candidate might wait long enough for an answer that the candidate took another offer. With one analyst now answering in real time, that kind of decision stopped stalling.
This is where productivity and strategic impact show up together. The productivity gain is headcount: one analyst doing work that used to require several. The strategic impact is what that analyst's speed did to the relationship between finance and the business, measured less by a number and more by whether VPs still bring decisions to FP&A first or learn to route around it.
Example 2: Identifying a New Market Segment
At the same company, my team had been focused on one segment, enterprise software, because that's where leadership had originally pointed us and nobody had the bandwidth to question it. Once AI cleaned up our data and shortened our reporting cycle, that freed-up time became the input for something we'd never had room to do: research whether another segment might convert better.
We used it to evaluate industrial and manufacturing buyers, a segment outside our usual focus.
Built an AI agent to pull buyer fit and propensity data for the new segment
Got a real read on demand in about a day instead of weeks
Tested the segment with actual partners before committing further
The research held up, and we moved from there to a sales motion and product bundle built around that segment. The productivity gain is the time we got back. The strategic impact is a revenue line that didn't exist in our plan a year earlier. To explain the value of that AI workflow, I’d point to the revenue line’s performance over the next quarters and years as the ROI, not just the hours saved on market research.
Play the Long Game When Measuring the True ROI of AI in Finance
Neither example above showed up as a number in the quarter it happened. The headcount I avoided registered right away. The candidate we kept and the market we opened took a year or more to become anything measurable, and even then, only after we built a sales motion around it.
I'd rather measure decision quality than decision quantity. A year is a long wait for proof, but it's also how long it takes when the change is structural rather than just faster reporting. Take the efficiency and productivity wins in the short term. But don’t ignore the signs of strategic impact that show up beyond the P&L over the long term.
If this kind of thinking is useful, I write about AI's evolving role in finance regularly in my newsletter — connect with me on LinkedIn, or send me a message in the F Suite Braintrust.
Join The F Suite to walk through AI adoption with peers who are navigating the same problems in real-time.