← All insights
AI for CFOs Article Nov 2026 · 11 min read

AI Budgets Are Outrunning Results. What CFOs Should Fund in 2027

By the CFOLogic team

An issue of AI for CFOs, a weekly newsletter. Subscribe →

Most finance leaders know what the company signed for AI licences this year. Far fewer can say what the business spent on AI in total, or what it got back. That is the backdrop to the FY2027 AI budget, and for most companies the AI line is going up again.

That pattern is not new. At the end of 2024, Bain Capital Ventures surveyed about 50 CFOs. Nearly four in five expected their AI spend to rise in 2025. At the same time, 71% were not yet using generative AI in their finance function. Use has spread a great deal since then. The gap between spend and measured results is still there.

This piece is about the budget line itself. For where AI works in finance today, start with our guide to AI in the office of the CFO.

What the surveys say about AI budgets and results

Read these side by side. The samples differ, so the pattern matters more than any single number.

SurveyWho was askedOn spendingOn results or control
Bain & Company, Apr 2026100+ CFOs; half at companies with $5bn+ revenue83% plan to raise company-wide AI spending by more than 15% over two years31% are satisfied with their AI outcomes
Gartner, Feb 2026300+ CFOs and finance leaders, Oct 2025Nearly 60% plan to raise finance AI investment by 10% or more in 2026Not covered
U.S. Bank, Fall 20261,000 US senior finance leaders, $100m to $5bn+ revenue51% say AI tool and platform spend ran over budget in the past 12 monthsNot covered
CloudZero (vendor survey), Aug–Sep 2026252 US CFOs and VPs of finance87% are raising FY2027 AI budgets61% already spend more than they can justify with measurable outcomes
Deloitte CFO Signals, Q3 2026200 North American CFOs, $1bn+ revenue53% review AI usage costs only after the fact74% take several days to several weeks to act on an overrun
BCG, Sep 20261,330 executivesAI spend rose from about 1.7% of revenue in late 2025 to 3.3%Nearly half of companies now report capturing value from AI
SVB, Feb 2026200+ CFOs at venture-backed companiesMedian AI spend expected to reach $50,000 in 202651% report ROI from AI spending

One caution. Most of these samples are large companies, so a $20M business should not set its plan from their averages. The venture-backed figures are a useful contrast. A median of $50,000 is a modest sum, and at that size the bigger risk is that nobody checks what the money buys.

Why AI spending runs ahead of results

Four reasons come up again and again.

The spend is spread out. BCG found that more than 80% of AI spending now sits outside the enterprise IT budget. Some of it is in marketing, some in operations, some on company cards. Deloitte found that 53% of large-company CFOs review AI usage costs only after the fact, through vendor bills or periodic reports. If the total is not visible, it cannot be compared with anything.

Only the tool gets funded. A licence is easy to put in a plan. The work that makes it pay is harder to see: rewriting the procedure, fixing the data, setting up review, and training the people who use it. Bain's April 2026 research advises CFOs to pay down "workflow debt" before deploying agents. In most plans, that debt has no line of its own.

Usage costs are treated as fixed. Seat licences are predictable. Much of the newer spend rises with use, because it is charged per model call, token or agent run. Deloitte found that 38% of CFOs cite limited insight into token usage and unit costs as a barrier to overseeing AI spend.

There is no baseline. If nobody measured hours, error rates or cycle time before the tool went in, there is nothing to compare with afterwards. In CloudZero's survey, 44% said AI spend was approved without reliable projections, and 47% said their board asks for ROI proof they cannot provide. Our piece on measuring AI ROI in finance covers four measures worth tracking.

What belongs in next year's plan

Plan one process at a time. For each process you intend to change, cover these lines.

LineWhat it coversHow to plan it
Licences and seatsNamed users, platform fees, AI add-ons from your ERP vendorFixed. Check renewal dates and minimum commitments.
UsageModel calls, tokens, agent runs, pages processedVariable. Expected volume times unit price, with a monthly cap and an alert before it.
Integration and dataRead-only connections, field mapping, cleaning master dataMostly one-off. Often internal time.
Process redesignThe new procedure, exception routes, sign-off pointsInternal time or outside help. Rarely planned for.
Review timePeople who check AI output before it is usedRecurring. Net it against the hours the tool saves.
Controls and evidenceAccess reviews, logs, audit evidenceRecurring. See AI internal controls.
MeasurementA baseline before, tracking afterSmall. Without it, no other line can be judged.

Two quick checks on a draft. If licences make up nearly all of it, the plan is probably missing the work that produces results. If a line cannot be tied to a named process, ask what it is for.

If your FY2027 numbers are already with the board, you do not need to reopen the total. The stages below can sit inside the approved figure.

Release the money in stages

Most finance teams already do this with capital spend. Approve an envelope for the year, then release it as evidence comes in.

StageMoney released forEvidence needed to move on
1. MeasureA time audit, one view of AI spend across all functions, a baseline for each target processA baseline for each process and a single total for AI spend
2. PilotOne process, read-only data, human review, a fixed decision dateResults against the baseline, on the company's own data
3. ScaleRollout of the processes that beat their baselineThe gain holds for a full quarter, and cost per unit of work is known
4. ReviewA half-year review of every AI lineA decision for each line to stop, keep or expand

The rule that matters most sits in stage two. A pilot that misses its criteria by the decision date stops, and its money goes back to the pool. Without that rule, pilots tend to run on and the spend grows each year. Our guide to designing an AI proof of concept in finance sets out a twelve-week shape.

Point some of the money at decisions

Gartner's March 2026 survey of 204 finance leaders found that 45% of finance AI investment leans toward productivity and 20% toward decision quality. Productivity work is a sensible place to start. The gains do level off once a task is fast, though, and Gartner notes that boards put more weight on investments that improve decisions and drive growth.

A practical step is to fund one decision-focused use case alongside the efficiency work. In a $10M to $50M business, that might be a sharper cash forecast, collections prioritization that goes beyond the aging report, or variance commentary a board can trust. Measure these in cash, margin or forecast error rather than hours. Our scoring method for AI use cases helps compare candidates.

What to show the board

A single page covers most board questions if it shows:

Total AI spend across the business, not only in finance, against plan
Spend and results for each process, against its baseline
Usage cost per unit of work, such as cost per invoice processed
Any overruns, and what was done about them
Which lines reach a stage gate next quarter

Deloitte found that nearly 80% of large-company CFOs say finance has a formal role in AI decisions, and 41% are required approvers. Where finance approves the spend, finance should own this page too. Our note on building a first real annual budget for FY2027 covers the rest of the plan.

What better budgeting will not fix

A tighter plan makes waste easier to see, but the causes sit elsewhere.

It will not fix a weak process. If the close runs late because accruals are missed or nobody owns the reconciliations, AI adds a step to a process that is already late. It will not make a product fit your data either. Two tools built for the same task can perform very differently on your own records, which is what the pilot stage is for.

Usage costs will still move during the year, so give each usage line a range and a cap and expect to adjust it. And no format replaces an owner. A line with no named person behind it tends to drift.

Frequently asked questions

How much should a company budget for AI in finance?

There is no reliable benchmark for a mid-sized company, so build the number from the processes you plan to change. Survey figures describe large companies. Gartner found that 47% of CFOs put 1% to 5% of finance technology spend into AI, and BCG puts total AI spend at about 3.3% of revenue for the companies it studied. Treat those as context only.

Why are AI budgets rising when results are thin?

AI budgets are rising because the costs are easy to approve, while the results depend on work that is usually unfunded. Process redesign, data work, review time and measurement are often left out of the plan.

What should an AI budget include besides licences?

An AI budget should also cover usage costs, integration and data work, process redesign, review time, controls and measurement. Plan each one by process, and treat usage as a variable cost with a monthly cap.

How do you control usage-based AI costs?

Control usage-based AI costs with a monthly cap for each process, an alert before the cap is reached, and one person who can pause usage the same day. Track cost per unit of work, such as cost per invoice, so you can see whether each unit gets cheaper as volume grows.

Is it true that 95% of AI projects fail?

No, the 95% figure is not a general failure rate for AI. It comes from an MIT NANDA report from July 2025, and it describes generative AI pilots that showed no measurable effect on profit and loss. Larger 2026 surveys are less stark: SVB found 51% of venture-backed CFOs report ROI from AI, and BCG found nearly half of companies now report capturing value. The MIT work did make one point worth keeping. More than half of generative AI budgets went to sales and marketing tools, while the biggest returns were in back-office work.

Who should own the AI budget?

Each process needs one named owner, with finance holding the consolidated view and the stage decisions. Deloitte's Q3 2026 survey found that 41% of large-company CFOs are already required approvers for AI decisions.

Where CFOLogic fits

CFOLogic works with CFOs and finance teams on AI built on top of the systems they already run, with read-only access by default and a person signing off on anything that leaves the team. For an FY2027 plan, that work starts with the time audit and a single view of AI spend, before any new pilot is funded. Through CFO Success Partners, we support the finance leader who has to run that process alongside everything else on their desk.

Book a savings review →

Related insights

Sources

Bain Capital Ventures, "AI and the Office of the CFO in 2025," 18 February 2025. Survey of about 50 CFOs on BCV's CFO Advisory Board, end of 2024. https://baincapitalventures.com/insight/ai-and-the-office-of-the-cfo-in-2025/
Bain & Company, "42% of CFOs plan to increase AI investment by over 30% within two years," press release, 13 April 2026. Survey of more than 100 CFOs globally; half at companies with revenue of $5bn or more. https://www.bain.com/about/media-center/press-releases/2026/42-of-cfos-plan-to-increase-ai-investment-by-over-30-within-two-yearsbain--company/
Gartner, "Gartner Research Reveals CFOs' Budget Plans Prioritize Growth Functions, Technology and AI in 2026," press release, 10 February 2026. October 2025 survey of more than 300 CFOs and finance leaders. https://www.gartner.com/en/newsroom/press-releases/2026-02-10-gartner-research-reveals-cfos-budget-plans-prioritize-grotwth-functions-tech-and-ai-in-2026
Gartner, "Gartner Survey Shows 45% of CFOs Say Their AI Investments Lean Toward Productivity, While 20% Say These Investments Lean Toward Decision Quality," press release, 20 July 2026. March 2026 survey of 204 finance leaders. https://www.gartner.com/en/newsroom/press-releases/2026-07-20-gartner-survey-shows-45-percent-of-cfos-say-their-ai-investments-lean-towwards-productivity-while-20-percent-say-these-investments-lean-towards-decision-quality
U.S. Bank, "CFO Insights Report: Fall 2026." Survey of 1,000 US senior finance leaders at companies with $100m to $5bn+ revenue, 5–27 August 2026. The over-budget question excludes respondents who do not track AI usage costs. https://stage5.usbank.com/corporate-and-commercial-banking/cfo-insights.html
CloudZero, "FY2027 AI Spend Survey," 2026. Survey of 252 US CFOs and VPs of finance, August–September 2026. CloudZero sells AI spend allocation software and wrote the survey. https://www.cloudzero.com/fy2027-ai-spend-survey-report/
Deloitte, "CFO Signals Survey, Q3 2026," published 21 October 2026. Survey of 200 CFOs at North American organizations with revenue of at least $1bn. https://www.deloitte.com/us/en/insights/topics/business-strategy-growth/3q-2026-cfo-signals-survey.html
Boston Consulting Group, press release, 30 September 2026. Survey of 1,330 CxOs and senior leaders. https://www.bcg.com/press/30september2026-ai-starting-to-pay-off-companies-generate-value
Silicon Valley Bank, "State of the VC-Backed CFO 2026." Survey of more than 200 CFOs at venture-backed companies, February 2026. https://innovation.firstcitizens.com/trends-insights/reports/state-of-the-vc-backed-cfo
MIT NANDA, "The GenAI Divide: State of AI in Business 2025," July 2025, as reported by Fortune, "MIT report: 95% of generative AI pilots at companies are failing," 18 August 2025. https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
AI for CFOs Published Nov 2026 · CFOLogic Insights
The newsletter

AI for CFOs, in your inbox every Wednesday.