AI projects are easy to announce and surprisingly difficult to value.
Deloitte’s Finance Trends 2026 research found that 63% of surveyed finance teams had fully deployed and were actively using AI, while only 21% reported clear, measurable ROI. EY’s 2026 Global DNA of the CFO survey found that just 21% of CFOs described their finance function’s AI preparedness as leading or advanced.
Those findings point to a basic problem: adoption is not the same thing as value creation.
And measuring the difference belongs squarely in finance.
Start with the baseline
Before an AI tool is introduced, measure the process it is supposed to improve.
- How many hours does the work take?
- How many people touch it?
- What does the process cost?
- How long does it take to complete?
- What error rate exists today?
- What service level or output volume does the team produce?
Without a baseline, “AI made us more efficient” becomes almost impossible to prove.
Time saved is not automatically ROI
Suppose an AI workflow saves a finance team 2,000 hours a year.
That sounds valuable. It may be valuable. But the financial result depends on what happens to those hours.
- Did the organization avoid hiring additional staff?
- Did it reduce outside consulting or temporary labor?
- Did the same team process more work?
- Did analysts redirect time toward higher-value decisions?
- Did the close get faster or the forecast become more accurate?
- Did service quality improve enough to affect revenue or retention?
Capacity created by AI is an opportunity. It becomes ROI only when that capacity produces an identifiable benefit.
Measure the full cost
The vendor subscription is often the easiest cost to see and the least complete.
A realistic AI cost model may include:
- software licenses and usage-based charges
- cloud and compute costs
- implementation and integration
- data cleanup and preparation
- security and governance
- testing and monitoring
- training and change management
- internal staff time
- ongoing support and model maintenance
This becomes especially important as AI scales. Deloitte’s Finance Trends 2027 research found that 60% of finance leaders expect to need more sophisticated AI cost-management practices through 2027.
Value has more than one form
Not every worthwhile AI project produces a direct headcount reduction.
I would measure benefits in at least four categories.
1. Cost and capacity
Hours saved, hires avoided, external spending reduced, or transaction cost lowered.
2. Speed
Faster close, faster analysis, shorter customer response times, quicker contract review, or faster decision cycles.
3. Quality
Fewer errors, better forecast accuracy, more consistent reporting, improved anomaly detection, or better documentation.
4. Risk reduction
Stronger controls, faster detection of unusual activity, better policy compliance, or improved visibility into contracts and commitments.
Finance should own the benefit ledger
Every significant AI investment should have a simple benefit ledger.
- Baseline: What did the process look like before?
- Investment: What did we spend to implement and operate it?
- Target benefit: What outcome did we expect?
- Realized benefit: What actually changed?
- Owner: Who is accountable for converting the new capability into value?
- Next decision: Expand, maintain, redesign, or stop?
This prevents a common problem: technology teams report successful deployment while finance waits for the economic result to appear somewhere else.
The weakest AI business case is “everyone else is doing it”
Competitive pressure matters. Experimentation matters. Some investments need to be made before perfect ROI can be known.
But uncertainty is not a reason to abandon financial discipline.
It is a reason to fund in stages, measure continuously, and increase investment as evidence improves.
The question finance should ask is not, “Did we deploy AI?”
It is: “What changed because we deployed it, and was that change worth what we spent?”
Sources: Deloitte Finance Trends 2026; Deloitte Finance Trends 2027; EY 2026 Global DNA of the CFO Survey.
