Skip to content
Hintelia Hintelia
Back to CFO Corner

CFO Corner

What finance expects from AI

Faster processes, lower costs and better decisions are different promises. Finance needs to agree which ones its AI investments are meant to keep.

Hintelia

7 min read

In this article

For many finance teams, the question is no longer whether to use AI. It is what they expect it to change.

That distinction matters. A finance team might use AI to produce a management report in half the time. The CFO sees a productivity gain. The board asks whether the report led to a better decision. The team expects time for more analytical work; a budget owner sees scope to reduce staffing. The tool could work exactly as intended and still leave someone disappointed.

Our argument is that finance can adopt AI without a shared view of what success should look like. Faster processes, lower costs and better decisions are often bundled into the same promise. They require different choices, and evidence of one does not establish the others.

The research gives reasons to take that gap seriously. Protiviti’s 2026 Global Finance Trends Survey reports that 77% of finance organisations employ AI, but only 35% rate themselves highly or moderately effective at measuring its return. Just 14% employ it under a defined strategy. Those figures do not tell us that finance leaders lack expectations. They suggest that adoption is further advanced than the discipline needed to turn those expectations into demonstrable results.

Four answers on the table

There are at least four plausible answers to what AI should change in finance. A CFO may want all of them. That does not make them the same investment.

The same work, faster. Shorter reporting cycles and less manual reconciliation are useful outcomes in their own right. In a March 2026 Gartner survey of 204 finance leaders, 45% said their AI investments leaned towards productivity, while 20% said they leaned towards decision quality. A separate Gartner study found that data extraction, payables and receivables automation, and report creation generally delivered expected value within nine to ten months. These are reported averages, not a deadline every project should meet.

A lower cost of finance. Time saved can help a team absorb more work without hiring. It can also support staffing reductions. Neither happens simply because a task takes fewer minutes. Costs depend on what work disappears, what review remains and what it costs to run the new system. In research by the Atlanta and Richmond Feds and Duke University, financial executives reported a negligible aggregate effect of AI on company headcount in 2025 and expected an effect close to zero in 2026. Larger firms expected reductions. This is evidence about whole companies, not a forecast for finance departments, but it cautions against treating productivity and job cuts as synonyms.

Better decisions. A cash forecast might give management earlier warning of a funding shortfall. Scenario analysis might change the timing of an investment. Here, the benefit depends on whether the analysis changes a decision and improves its outcome. Gartner argues that boards place greater weight on growth, decision-making and competitive advantage than on the efficiency gains many finance AI investments pursue. A faster report may help, but speed alone does not answer that expectation.

A different way of running finance. BCG describes a future of continuous reconciliation, dynamic forecasting and more autonomous transaction processing, with people concentrating on exceptions and judgement. That is an ambitious operating model, not a result established by widespread AI use. It requires decisions about authority and controls: what the system may do, what a person must approve and who is accountable when it is wrong.

Where the expectations collide

The problem begins when these answers are treated as interchangeable.

Suppose an AI tool halves the time needed to prepare a forecast. That demonstrates a process improvement. It does not yet show that the forecast is more accurate, that management acts on it sooner or that finance costs have fallen. Each claim needs its own evidence.

The same applies to the hours saved. If the budget assumes those hours will reduce costs while the business case assumes they will be spent on deeper analysis, the benefits may be counted twice. The CFO needs to decide how the capacity will be used.

There is evidence that people anticipate different workforce outcomes. In Firm Data on AI, a study surveying almost 6,000 executives across four countries, executives expected AI to reduce employment at their firms by 0.7% over the following three years. A separate employee survey expected a 0.5% increase. These are different respondent groups, not proof that managers and staff disagree inside every company. They do show why the consequences for jobs should be discussed rather than left implicit.

Even reported returns can conceal uncertainty. CFO Dive’s coverage of Gartner’s research found that 55% of CFOs reported positive overall returns from their 2025 AI initiatives, yet 57% said returns from individual use cases were unclear. A favourable view of the portfolio does not tell a CFO which initiative deserves the next investment.

The studies cover different populations, periods and definitions of AI. Some concern finance functions; others concern entire businesses. They cannot be combined into a single measure of success. Taken together, they support a narrower conclusion: reported adoption, expected benefits and demonstrated value are different things.

Where we stand

We favour using AI to strengthen the decisions finance helps the business make. Saved time creates room for that work; it does not guarantee it.

That is our preference, not a result the surveys prove. Finance earns its place in business decisions by testing assumptions, explaining consequences and challenging the use of capital. We would judge its broader AI programme by whether it does that work better.

We would still fund an efficiency project that delivers a worthwhile return. A cheaper, reliable reconciliation process does not need to become a strategic transformation to justify itself. Nor should a forecasting initiative be judged only by the time saved producing the forecast. Each deserves a business case suited to its purpose.

For a decision-focused investment, name the decision first. If the aim is better liquidity management, assess forecast error at a useful horizon and how much earlier the team identifies a funding need. Agree who will act on that information. Producing more scenarios is an activity; improving the choice between them is the ambition. Business outcomes also depend on events outside finance’s control, so judging decision quality requires more than attributing every subsequent gain or loss to AI.

For an efficiency investment, measure the whole process, including checking and correcting the output. Distinguish capacity released from cash actually saved. If people are expected to spend more time with the business, their priorities and workload need to change. If the plan instead assumes fewer roles, make that explicit and test whether the work can be done reliably before removing the capacity.

Throughout, finance must retain an answer to a familiar question: can we explain and stand behind these numbers? Access to source data, clear approval responsibilities and proportionate human review belong in the design. An output that arrives quickly but cannot be checked may add work and risk rather than remove them.

Before approving an initiative, the CFO, its business owner and the people doing the work should agree what will change, how they will recognise an improvement and when they will review the evidence. Experiments can start with uncertainty. They should still have a question to answer.

A faster process can be a success. A better decision can be a success. Finance needs to agree which it is buying before it claims to have delivered both.

Sources

Hintelia

You bring the ambition; we bring the structure.

Contact us