AI ROI in Finance: Closing the Gap Between What You're Spending and What You're Seeing

66% of CFOs expect significant AI ROI within two years — yet only 14% report meaningful value today, according to RGP's December 2025 survey of 200 U.S. finance leaders. If you have sat in a budget review defending AI spend against a blank line in the value column, you are not alone, and you are not failing. You are navigating a structural gap that is reshaping every finance function in enterprise America right now.

The tension is real. The ambition is not wrong. But the path from here to there requires something most AI roadmaps never address.

What conventional advice gets wrong

The standard prescription goes something like this: identify a high-value use case, run a pilot, measure the lift, scale. It sounds reasonable. It has produced almost nothing.

Gartner's survey of 183 CFOs found that 91% of finance organizations report low or moderate AI impact in early stages of adoption. Twenty-five percent remain uncertain how to move from planning to piloting at all. The conventional advice assumes the foundation is ready. It is not.

RGP's research names the actual blockers plainly: fragile data foundations, legacy systems, and widening workforce capability gaps. These are not gaps that a new AI vendor fills. They are gaps that a new AI vendor reveals. Buying a more sophisticated tool and placing it on a broken data architecture is not a pilot — it is an audit of your pre-existing problems, billed at enterprise rates.

The other thing conventional advice gets wrong is the measurement frame. Finance leaders are being asked to demonstrate ROI on AI investments using the same quarterly reporting logic they apply to capital expenditures. But AI capability compounds differently. The first 90 days often produce negative visible ROI — you are cleaning data, training people, rebuilding workflows. The return lives in month seven, month fourteen. If you are measuring with the wrong clock, you will kill the right investment.

What Marcus Aurelius sees in this

In Meditations Book V, I wrote: "Confine yourself to the present." Not as an excuse for passivity — but as a discipline of attention. The present is the only place where work actually happens. Most AI programs in finance are living everywhere except the present: in the ambition of what AI could do, in the anxiety of what competitors might be doing, in the politics of what leadership expects to see by Q4.

This reveals a failure not of technology, but of philosophy. The CFO who cannot show AI ROI today is often suffering from what the Stoics would call phantasia — the uncritical acceptance of an impression. The impression here is that AI is a procurement decision. You buy it, you deploy it, the value appears. This impression, accepted without examination, drives the gap between the 66% who expect results and the 14% who have them.

Epictetus taught in the Discourses that we must distinguish between what is eph' hēmin — up to us — and what is not. The market conditions that created this AI moment are not up to you. The vendor's product roadmap is not up to you. But the integrity of your data is up to you. The capability of your team is up to you. The clarity of the problem you are trying to solve — that is entirely up to you.

This means the ROI gap is, at its root, a governance problem wearing a technology costume. The organizations reporting meaningful AI value today did not find better tools. They did the unglamorous interior work first: they audited where their data broke down, they identified one workflow where the input and output were already clean, and they built from there. They practiced what I would call prosoche — sustained attention to what is actually happening, not what is hoped for.

The examined work life demands that you stop measuring AI against the fantasy you bought and start measuring it against the process you actually run. Book IV of the Meditations puts it directly: "The impediment to action advances action. What stands in the way becomes the way." Your legacy systems are not obstacles to your AI strategy. They are the strategy — the exact terrain on which real, measurable value must be built or not built at all.

The practical path

Start with one process that already has clean, structured inputs and a measurable output. Month-end close is a strong candidate. Variance analysis is another. These are not the most exciting AI applications in finance. They are the ones where you can actually demonstrate a before and after — processing time, error rate, hours recovered.

Use that demonstration to build the internal case for the harder infrastructure work. Data governance is not an IT project. It is a finance leadership project. You own the numbers. You must own the quality of the numbers.

Build capability in your team in parallel, not sequentially. Do not wait until the platform is ready to start training your analysts. The AI Cost Modeling for Finance Leaders framework is a useful starting point — it reduces planning time and, more importantly, it gives your team a concrete, bounded problem to develop AI fluency around.

When you are ready to expand, work outward from the close process into forecasting. Multi-Scenario Cash Flow Forecasting gives you a structured approach to building scenario models that your business partners can actually interrogate — which is where the stakeholder value begins to compound.

Measure three things, not one: time recovered, error reduction, and decision quality. ROI in AI-augmented finance is not a single number. It is a pattern across multiple leading indicators that, together, tell the business what the investment is producing.

What to do this week

Before you close this tab, open the calendar and block two hours this week — not for a strategy meeting, not for a vendor call. Block two hours to walk one finance workflow from raw data input to final output and write down, by hand, every point where data quality degrades, where manual intervention occurs, or where a colleague is performing a task that exists solely to compensate for a broken upstream process.

One workflow. Two hours. A written list of what you find.

That list is your AI roadmap. Not the vendor deck. Not the pilot proposal. The specific, documented friction in the specific process you actually run. Build from that list and you will be in the 14% — and ahead of them — by the time the rest of the market finishes its planning cycle.

The Design AI-Enhanced Month-End Close Workflow prompt can help you translate that list into a structured implementation path before the week is out.

Explore further

If the close process is where you are starting, Build a Three-Statement Financial Model That Actually Balances walks through the structural integrity your model needs before automation adds value rather than amplifying error.

For communicating AI impact to executive stakeholders — which is its own distinct skill — AI-Enhanced Financial Storytelling: Engage Stakeholders addresses how to translate operational metrics into the language that earns continued investment.

And if payroll is a process carrying disproportionate manual burden in your function, AI Payroll Processing | Reduce Processing Time by 75% is worth reviewing as a contained, measurable first win.