The numbers tell a striking story of rapid adoption and disappointing returns.
According to Gartner's 2025 survey of 183 CFOs and senior finance leaders, 59% now report using AI in their departments — up from 37% just two years earlier. The share of finance teams using AI tools jumped from 34% to 72% in a single year, according to separate research.
Yet only 7% of CFOs report high ROI from AI in finance functions. A survey of 200 U.S. finance chiefs by professional services firm RGP found that only 14% have seen a clear, measurable impact from their AI investments to date. Deloitte research paints a similarly sobering picture: only 21% of active AI users in finance say it has delivered clear, measurable value.
Worldwide spending on AI is forecast to reach $2.52 trillion in 2026 — a 44% increase year over year. The investment is accelerating. The returns, for most, are not.
This is not a technology failure. It is a strategy failure.
Why AI Investments Underperform in Finance
The Solution-First Trap
The most common pattern in finance AI adoption is technology-led implementation. A vendor demonstrates an impressive capability — automated reconciliation, predictive forecasting, natural language report generation — and the organisation procures it, deploys it, and searches for problems it might solve.
This inverts the correct sequence. Effective AI deployment starts with a specific, measurable business problem — one where the cost of suboptimal decisions is quantifiable — and works backward to the appropriate technology.
Gartner's research confirms this pattern. Knowledge management — helping organisations organise, retrieve, and leverage information — is the most common AI use case in finance at 49%, followed by accounts payable process automation at 37%, and error and anomaly detection at 34%. These are legitimate use cases. But when they are adopted without a clear baseline measurement and decision-support framework, the investment cannot be evaluated, and the value cannot be captured.
The Data Foundation Gap
AI models are only as good as the data they consume. The same data quality challenges that undermine FP&A frameworks — fragmented architectures, inconsistent definitions, manual data handling — are amplified when machine learning models attempt to extract patterns from unreliable inputs.
Finance functions that rush to deploy AI before investing in data governance are building on sand. The models may be sophisticated. The outputs may be wrong.
Before investing in AI capabilities, finance leaders should invest in the prerequisites: consistent data definitions, reliable pipelines, clear data ownership, and a semantic layer that ensures the same term means the same thing across the organisation.
The Change Management Deficit
An AI model that produces better forecasts creates no value if the planning process does not change to use those forecasts differently. Technology adoption without process redesign is waste.
Consider a common scenario: a finance team deploys an AI-powered forecasting tool that demonstrably reduces forecast error by 15%. Impressive. But if the planning calendar, review process, and decision-making protocols remain unchanged — if leadership still reviews a static monthly report two weeks after month-end — the improved forecast accuracy creates no incremental value. The bottleneck was never the forecast. It was the process.
The Talent Bottleneck
AI adoption requires a combination of technical competence and domain expertise that is rare in most finance functions. Data scientists lack finance domain knowledge. Finance professionals lack technical fluency. The gap between them creates AI implementations that are technically sound but operationally irrelevant.
Gartner's findings suggest that AI adoption momentum in finance has slowed — from the sharp increase between 2023 and 2024, adoption has essentially plateaued. One key reason: organisations that made early investments are now confronting the difficulty of scaling pilot projects into enterprise-wide capabilities. The technology works. The organisation is not yet ready.
A Practical Framework for AI ROI
Step 1: Start with the Decision, Not the Technology
What specific decision will this AI capability improve? How is that decision made today? What is the cost — in time, accuracy, or missed opportunity — of the current approach?
If these questions cannot be answered with specificity, the AI initiative is not ready for investment. It is ready for further problem definition.
Step 2: Quantify the Baseline Before Deployment
Before deploying AI, measure current performance rigorously. Forecast accuracy. Processing time. Error rates. Decision cycle time. Whatever metrics define success for the use case.
Without a baseline, there is no way to measure impact. Without measurable impact, there is no way to calculate return. And without return, AI spending is an act of faith, not an investment.
Step 3: Design the Process Change Alongside the Technology
AI is an input to a process, not a replacement for one. For every AI deployment, design the new process explicitly: who receives the AI output, what decisions they make with it, how those decisions are escalated, and how outcomes are tracked.
The process redesign often delivers more value than the AI itself. An AI-augmented planning process that reduces decision cycle time from two weeks to two days creates value regardless of whether the AI model's accuracy improvement is 5% or 15%.
Step 4: Measure Incrementally and Honestly
Track the marginal improvement that AI delivers against the baseline. Not the total value of the process — the incremental value of the AI component. Be honest about what the technology contributed and what the process redesign contributed.
This discipline prevents the common trap of attributing all process improvement to AI — which inflates expectations and distorts future investment decisions.
The Path Forward
Despite the current ROI gap, optimism is not misplaced. Two-thirds of CFOs using AI are more optimistic about its potential than they were a year ago. The technology is maturing rapidly. The use cases are becoming clearer.
But optimism without discipline produces the same outcome as pessimism: no measurable value.
The CFOs who will capture genuine returns from AI are those who treat it as a business investment — with clear problem definitions, measurable baselines, designed process changes, and honest impact assessment. The technology is not the bottleneck. Strategy is.