AI and AI agents will no doubt do a lot of work in a lot of industries in the very near future. But generative AI still carries the risk of creating incorrect or fabricated information.
In the world of finance and money movement, that's a big drawback. Finance needs AI that is trustworthy and, if need be, is checked by a human before decisions are made — not AI that tries to do it all.
For now, with AI agents still in their infancy, finance teams need to value predictability and correctness more than AI autonomy. They'll prefer boring precision over untrustworthy creativity and AI that works all of the time on a narrow task to AI that works 80% of the time on a broad task.
Trust and accuracy are goal 1
Teams rely on accuracy. They demand information that they can audit — and correct if it is wrong — as well as explain and rely on every time.
Generative AI capabilities are not quite there. Ask ChatGPT something in one way and it'll produce an answer that differs if it is asked the same question in a different way. This might not matter to a marketer creating a marketing plan or even a writer editing content. But for finance, a tool that occasionally outputs nonsense or different results is a non-starter.
Consistency isn't just a technical preference — it's an adoption blocker. In short, if an AI's answers can't be predicted and justified, CFOs won't put it anywhere near the books.
This doesn't mean AI isn't making inroads into finance. Market researcher Gartner found that 58% of finance leaders had adopted AI in 2024, up 21 percentage points from 2023. The biggest use cases? Low-risk ones, including leveraging capabilities of existing automation tools and anomaly and error detection. Only 28% used it for the creation of better financial forecasts and results analysis feeding decision making.
Deterministic vs. probabilistic
Professionals will continue to be cautious in adopting AI because, at the core, the new wave of large language models are probabilistic, not deterministic. In contrast, CFOs are accustomed to software that behaves predictably. Given the same inputs, it produces the same outputs every time. Such traditional rule-based systems are deterministic and never deviate unless explicitly reprogrammed. CFOs expect the same level of predictability with AI agents, and AI developers in finance are finding ways to combine the flexibility of AI with the guardrails of deterministic logic. For example, techniques are emerging to ensure that teams can trace and verify every AI's decision.
That said, finance leaders are still mostly investing in narrow AI use cases that create immediate returns because they are:
- Highly repeatable: By automating well-defined tasks, AI can take up some of the boring work that finance workers get stuck doing. Accounts payable processing is one such example. AI can not only draft vendor bills, but it can also predict general ledger codes for each invoice. This replaces manual coding, saving companies time and money. Similarly, expense management tools now use AI to automatically categorize transactions and flag anomalies.
- Involve highly prescribed workflows: AI agents stick to highly prescribed workflows that provide a set, repeatable structure. This means they do exactly what they're designed to do, and they do it every time. This is the current state of AI agents, a sort of level one in terms of development. It might seem underwhelming when compared to loftier claims and aspirations for AI in the future, but agents that do exactly what they're supposed to do, over and over, is music to the ears of a CFO. It means the technology works consistently.
- Involve humans: AI can take some initiative but only under strict constraints and human oversight. AI doesn't replace a finance team's judgement, it frees the team up to do higher level work.
Not disrupt; quietly make better
To evaluate an AI solution, finance leaders should insist on determinism and control and look for guardrails and guarantees. All the while, they need to keep predictability, auditability and compliance front and center.
For AI in finance, the goal is not to disrupt everything, but for AI to quietly make finance operations better by catching errors, saving time, and never dropping surprises.
