A couple of years ago, most writing about AI in finance was a forecast. CFOs were told that predictive models, agents and real-time dashboards would reshape the function, and that the open question was how fast. Enough time has passed to check those forecasts against what finance leaders actually report doing.
This post looks at the question from the CFO's chair: where the AI budget is going, which early expectations held up, which did not, and why trust has turned out to be the constraint that sets the pace. Every adoption figure below comes from a named public survey, cited in the sentence.
How many finance teams actually use AI?
Roughly six in ten, and that share has stopped climbing quickly. Gartner reported that finance AI use jumped from 37 percent in 2023 to 58 percent in 2024. Its next survey, of 183 CFOs and senior finance leaders, found 59 percent in 2025, essentially flat.
The plateau is the useful signal. The teams that wanted a first AI tool got one early. The remaining group is split between leaders still moving from planning to a pilot and a smaller group with nothing planned. For the CFOs already using AI, the question has shifted from whether to adopt to whether the tools they bought are paying back.
Which early expectations held up?
Automation of repetitive work came first
The prediction that AI would start with high-volume, rules-heavy work was right. In Gartner's 2025 survey the most common uses were knowledge management (49 percent of adopters), accounts payable automation (37 percent) and error and anomaly detection (34 percent), as reported by CFO Dive. These are tasks with clear inputs, clear right answers and a human who can check the output.
Data quality turned out to be the gating factor
Early writing warned that AI is only as good as the data feeding it. That warning aged well. Finance leaders in the same Gartner survey named weak data literacy and technical skills as the largest obstacle, with inadequate data quality or availability close behind. Teams with ERP, billing, CRM and bank data in separate places spend the first months of any AI project on integration, before any model does useful work.
Risk detection proved a practical early win
Spotting anomalies in payments, spend and customer behaviour was expected to be a strong use case, and anomaly detection is now among the three most common uses Gartner found. It suits the way finance already works: the system flags, a person decides.
Which expectations did not hold up?
Fast, broad impact
Many forecasts implied that AI would change finance outcomes quickly. The survey data describes a slower curve. Gartner's 2025 survey found that 91 percent of adopters saw low or moderate impact at first, and that teams further along were several times more likely to report high impact. Value arrives with maturity, which takes time and repeated use on real data.
Productivity as the whole story
Most finance AI spending still targets productivity. In a survey of 204 finance leaders, Gartner found that 45 percent of CFOs say their AI investments lean toward productivity and 20 percent toward decision quality. Coverage of the same research reported that CFOs focused on decision quality were more likely to describe the value as significant, 31 percent against 17 percent for productivity-focused investments. The early pitch treated time saved as the payoff. The data suggests the larger return comes when AI changes what the team decides.
Split-second, dashboard-driven decisions
The picture of a CFO repricing mid-quarter off a real-time AI dashboard has not become the typical pattern in the surveys. Dashboards and forecasting help, but the most common uses remain operational. Strategic decisions still move at the speed of review cycles and board calendars.
Where are CFOs putting AI budget now?
Three patterns show up in the public research.
First, spending is concentrating on processes with measurable cycle times, such as payables, reconciliation, collections and cash application, where a CFO can compare before and after. Second, CFOs are asking vendors to prove value on their own data before signing, because a demo on clean sample data says little about their own ERP. Third, budget is moving from tools that execute a single step toward systems that make judgement calls across a process: deciding which accounts to prioritise, what to say to a customer, and when to escalate.
In 2024, Gartner predicted that 90 percent of finance functions would deploy at least one AI-enabled solution by 2026. Whether or not that exact figure lands, the more useful question for a CFO is how deep each deployment goes, since one tool in one step rarely moves working capital.
Why is trust the real constraint?
Handing a task to software that makes decisions feels different from handing it to software that follows rules. A CFO signs off on the numbers and answers for them, so the bar for delegation is high, and rightly so.
The finance teams that have moved furthest tend to build trust in stages:
- Start with a process where every AI output can be reviewed before it takes effect.
- Measure accuracy against the team's own history, on the team's own data.
- Expand the scope of what the system decides only after the review rate shows it is reliable.
- Reassign people toward exceptions, customer conversations and analysis as the routine volume moves, and decide on capacity once the results are in.
Transparency matters as much as accuracy. Leaders want to see why the system made a call, what data it used and who can override it. Governance questions about data privacy, bias and workforce impact have not gone away, and a staged rollout is how most teams address them in practice.
What should a SaaS CFO take from this?
Key takeaways for finance leaders weighing their next AI investment:
- Adoption has levelled off at around six in ten finance functions, so the competitive question now is depth of use.
- Data readiness and team skills set the pace more than model capability does.
- Productivity gains are real, but the reported value is higher when AI improves decisions.
- Trust is earned through staged delegation, reviewable outputs and proof on your own data.
For SaaS finance teams, order-to-cash is a natural place to apply these lessons: it is high volume, it runs on data the team already owns, and the decisions in it, such as which invoice to chase, which customer is likely to slip, and how cash will land, show up directly in DSO and the cash forecast.
If you want to see what an AI analyst that makes those calls across the order-to-cash cycle looks like on your own data, see how Tesorio's AR acts.
