The finance function is still struggling to trust artificial intelligence. However, without trust, the impact of AI remains weak, and the results are scarce.
Most finance teams are already using AI. But the time for experimentation is over: boards and investors now expect measurable results. However, the impact of AI still seems limited. According to a Gartner study, 91% of financial managers declared in 2025 that they saw only a “low or moderate” impact of AI on their real performance indicators.
The reason? Teams still struggle to rely on AI results in a reliable and verifiable way. Behind this gap between adoption and impact, there is only one obstacle: trust.
Why trust is slow to build
Finance is based on traceability. Each figure must be linked to a source, pass through systems and be able to be justified at any time, particularly during an audit. Decisions require precision, and results must constantly be evaluated and validated. But teams struggle to integrate AI results into processes like billing, revenue recognition, and closing, and often can’t explain how the results were generated. But in finance, a result that cannot be explained is a result that cannot be used.
Much of the problem comes from how AI is deployed. In many companies, it still operates outside the reference system: data is extracted, processed in external tools, and then re-integrated, which creates gaps in the audit trail and makes it difficult to track results across the entire quote-to-cash cycle. Concerns are constant: data security, lack of supervision, quality and reliability of the information processed.
Added to this is the question of context: the quality of the results generated by AI is directly conditioned by the quality of the data it receives. Even seemingly uncomplicated areas like taxation require specific context: place of residence, dependents, deductions and other particularities. Without this level of detail, results lack consistency and confidence erodes.
At the same time, there is also a deeper risk: if AI replaces critical reasoning instead of fueling it, teams will gradually lose the reflex to question the results.
How to build AI that finance can trust
Trust is therefore built from solid foundations. For AI to be truly reliable in finance, it must not be conceived as an additional technological layer added to existing systems. It must operate on complete data, within established controls, and produce results traceable back to the underlying transactions. Financial reference systems reflect years of accumulated business logic, compliance requirements, and special cases. When AI operates within this framework, it inherits the context and is naturally less exposed to risks.”
Trust is also based on an explicit validation process. Teams need to know exactly what controls to exercise before acting on an outcome, whether it’s a predefined checklist, mandatory criteria, or clearly established escalation procedures. The challenge is not to eliminate these controls, but to make them evolve to respond to a new environment.
AI in finance is entering a pivotal period. The question is no longer whether teams use it, but whether its results meet the audit and compliance standards that govern the entire finance function. Until this is the case, the gap between adoption and real impact will persist. Productivity matters, of course, but in finance, it is trust that determines what truly transforms the finance function, and what remains unheeded.
