AI is creating a lot of excitement in finance, but many finance leaders are still struggling to turn that excitement into measurable ROI.
Many teams are investing time and money into AI projects, then judging the results against measures that don’t fit where the technology is today.
BCG’s 2025 finance research found median reported ROI from AI at just 10%, with nearly a third of finance leaders saying they had seen limited or no gains. Gartner has also warned that a significant share of generative AI projects will be abandoned after proof of concept because of poor data, weak controls, rising costs or unclear business value.
So when I say most AI projects will deliver zero ROI, I don’t mean that AI lacks value. I mean a lot of businesses are still measuring the wrong thing.
For most finance teams, 2026 is a year for improving how work gets done. The bigger leverage comes later, once the capability, trust and control environment are stronger. That is the progression I keep coming back to in my work with finance teams: efficiency first, then leverage.
The ROI lens is the problem
Too many finance AI projects are still being judged on questions like:
- Can it replace part of the team?
- How much money did this save us?
- Can it run the task end to end?
Based on where AI is today, these expectations are simply too big, too early, and can send leaders in the wrong direction. That is where the gap often appears. If people expect AI to own tasks entirely, return won’t land.
A better starting point is much more practical: what finance outcomes are we trying to improve?
That might be a shorter close, cleaner reporting, better forecast preparation, fewer manual handoffs, or more finance time spent supporting decisions. Those are far more useful measures, and they fit much better with where most teams actually are.
Time saved is only part of the story
Efficiency matters. Finance teams are stretched, so any reduction in manual effort is worth paying attention to.
But time saved on its own is only the start. The bigger question is what happens with that time.
Does review quality improve? Are issues picked up earlier? Is there more space for business partnering? Are decisions getting better?
That is the line finance leaders need to draw more clearly.
- Activity gains are things like faster drafting, less formatting and fewer manual steps.
- Business gains are things like better forecast quality, shorter cycle times, fewer errors and more finance capacity for decision support.
If the time saved never gets reinvested in something meaningful, the return stays small and hard to measure.
The best use cases improve the operator
Right now, the strongest AI use cases in finance are the ones that improve how people work rather than trying to replace judgement-heavy work altogether.
I often frame this as the difference between “what” problems and “how” problems. A “what” problem asks AI to do the work. A “how” problem uses AI to improve how a person does the work. For most finance teams today, the better returns are still coming from the second category.
Good use cases are the “how” problems. Things like:
- Variance commentary support
- Scenario prep before forecast discussions
- Reporting pack preparation
- Identifying process gaps
These are practical, repeatable and easier to review. They can save time, improve consistency and reduce frustration without asking AI to own high-risk judgement.
By contrast, fully autonomous decision-making in sensitive reporting areas is still a poor place to hunt for ROI. The controls, trust and data quality usually are not there yet.
What should count as ROI in 2026?
For this stage of AI, I would measure ROI more broadly and more honestly.
Time saved should still be part of the picture. But on its own, it’s too narrow. If a team saves time without improving decision-making, process quality or finance capacity, the return is usually smaller than it first appears.
So for 2026, a more realistic way to measure ROI might look something like this:
- Time saved and cycle-time improvement
- Error reduction
- Process consistency
- Increased capacity for business partnering
- Earlier visibility of risks
- Better decisions, not just faster outputs
That is much closer to where value is actually being created right now. It also gives finance leaders a better filter for deciding which AI ideas are worth backing.
A practical framework for choosing the right use cases
Once ROI is defined properly, the next step is choosing projects that have a realistic chance of delivering it. Before backing an AI use case, score it from 1-5 against six questions:
- Frequency: how often does this task happen?
- Time intensity: how much effort does it genuinely consume?
- Decision impact: if this improved, would the business feel the difference?
- Data readiness: is the underlying data good enough to support it?
- Control and risk: can this be reviewed safely and responsibly?
- Workflow fit: can this slot into how the team already works?
The point is to stop choosing projects based on how polished the demo looks and start choosing them based on how much real value they can create.
The best early use cases are usually the ones that are frequent, time-heavy, easy to review and close to a real pain point in the workflow. They may not sound as exciting as fully autonomous finance, but they are much more likely to produce measurable value.
Adoption is part of the economics
Even a well-chosen use case can underperform if adoption is weak.
In finance, return comes from repeatability. If AI becomes part of the weekly rhythm of reporting, planning or analysis, the gains start to build. Teams learn what good looks like, confidence improves, and the output becomes more useful over time.
Readiness matters here. A team that can prompt well, review properly and challenge outputs sensibly will get more value from a simple use case than an unprepared team will get from a more advanced one.
2027 should be a different conversation
I’m optimistic about where AI is going.
As the tools improve, and as security and integration catch up, AI will become more strategically useful across finance. Conversations will be able to shift towards stronger scenario modelling, earlier risk identification and more forward-looking support to the business.
In the meantime, finance leaders should focus on getting prepared. Start by separating experimentation from real use cases you expect to scale. Then choose a small number of opportunities that are frequent, measurable and tied to a genuine finance pain point.
Measure success more carefully. Go beyond hours saved and look at review quality, cycle time, consistency and whether the team is creating more space for decision support.
Then invest in team readiness. Share good examples, compare prompts, review outputs together and make sure someone owns adoption after the pilot.
The teams that see strategic ROI fastest will usually be the ones that used 2026 well.
Ready to make AI deliver ROI in finance?
If you are a CFO, FD or finance leader trying to work out where AI can genuinely improve performance, this is exactly the work I support clients with. If you want to explore what that could look like in your business, get in touch.
