Opinion
AI can produce the answer. It can’t own the decision.
As AI produces forecasts, reports and analysis in seconds, the critical skill is no longer generating the output. It is knowing which assumptions to challenge before the business acts on it.
A finance leader can now ask AI to build a forecast, explain a variance and draft the narrative for the next board meeting. Within minutes, all three arrive looking polished, detailed and entirely convincing. The problem is that they can also be wrong in exactly the same way: each may rest on assumptions that no longer reflect the business.
That is why experience, institutional knowledge and human judgment matter more, not less. When an answer does not pass the smell test, the most important skill is knowing where to challenge it.
For years, professional expertise was often demonstrated through mastery of processes. The person who understood the system, knew where the data lived and could produce the correct report on time held knowledge that few others possessed. Accessing the information, structuring it and turning it into something useful required hard-earned experience.
AI has leveled that playing field. But it has also introduced new risks that can trip up even seasoned finance leaders.
A forecast is never simply a collection of numbers. It contains assumptions about demand, pricing, hiring, customer behavior, payment timing, operating costs and the choices management is likely to make. An AI tool can calculate the effect of those assumptions perfectly and write a persuasive explanation of the likely result. Yet a technically correct calculation can describe a state of affairs that no longer exists. A sales forecast may rely on a conversion rate that was valid before the market shifted. A healthy cash projection may conceal an emerging trend of customers paying later, even though no individual account looks alarming.
The figures in these cases are not necessarily false. But their significance has changed, and recognizing that change requires more than running the process correctly.
Finance teams have used rules, software and automation for decades. AI has now expanded what an organization can see: a deviation that hasn't yet appeared in a standard report, a risk hidden inside a familiar assumption, or a question nobody thought to ask. Greater visibility also creates a higher standard for professional judgment. Finance teams gain stronger analytical tools, while management faces the growing risk of treating a plausible (but mistaken) output as a conclusion the business can safely act upon.
A plausible explanation based on an outdated assumption can travel much further because it looks reasonable to anyone who lacks the context to question it.
At the same time, AI increases the risk of silos. An analyst can go deep in a single AI session, chasing an interesting thread, and lose sight of how that work connects to what the rest of the department is measuring and doing. Running a forecast across the entire finance function is a different proposition than running it alone, which is why an organization must preserve the definitions of its metrics, record changes to assumptions, apply the appropriate permissions, and reproduce analyses whenever management, the board or an auditor asks how the result was reached.
In fact, we've just surveyed 270 enterprise CFOs and found that 75% say the inability to audit or explain an AI-generated output is what's stopping them from trusting their tools with the work that matters most.
Human judgment can be effective only if everyone is working from the same data, definitions, assumptions and permissions. That is increasingly the job of a finance operating system, a governed layer that keeps data consolidated and contextualized so any AI tool draws on the same trusted source, rather than each session reconstructing its own version of the truth.
The underlying financial logic isn't fixed. New cost centers appear. General-ledger mappings are revised. Intercompany eliminations change and allocation methods evolve. Even the way a company defines an important metric may be updated. A result cannot support a responsible decision if every new conversation begins by reconstructing that logic from memory.
Take a model that attributes declining profitability only to seasonal costs. The explanation may fit the historical data, while a sales manager recognizes that customer behavior has changed due to heavier discounting to close deals in a slower market. The original calculation may be accurate within its frame of reference, but acting on it could still lead the company in the wrong direction.
The person who catches the problem may not be the one who knows the system best. It may be the one who understands the facts on the ground, notices that a pattern has changed, and asks what the existing forecast was never designed to explain.
AI is separating two activities that were once closely connected: producing professional work and understanding what that work means. Companies will need finance teams that can challenge an assumption, connect a financial signal to operational realities, and explain the consequences of every choice. Organizations should automate repeatable activity while giving finance professionals more capacity to investigate anomalies, test comfortable explanations, and advise the business on what to do next.
AI will continue to produce better analysis and may identify patterns that no individual would have found. But before an output becomes a business decision, someone must understand what it rests on, explain how it was reached, and accept responsibility for its consequences.
Didi Gurfinkel is co-founder and CEO of Datarails.
