AI cost controls lag as CFOs' influence grows, Deloitte finds
CFOs are taking a bigger role in AI investments, but spending controls aren't keeping pace. Deloitte found 53% still review AI usage costs after the fact.
AI spending is testing CIOs' ability to keep costs in check.
Usage-based charges and shifting vendor terms make AI costs hard to predict. Still, 53% of CFOs surveyed said their organizations primarily review AI usage costs retrospectively, through bills and reports, rather than monitoring spending in real time, according to Deloitte's CFO Signals survey, released this week. That can allow spending to exceed budgets before leaders spot the problem.
This challenge is drawing finance leaders further into enterprise AI investment and cost oversight, a shift highlighted in Deloitte's Finance Trends 2027 report released last month.
"The CFO is playing a more prominent role beyond the financial statements themselves and the finances of the organization," said Ed Hardy, U.S. finance services leader at Deloitte. "They're being pulled more into strategy."
Finance leaders are helping shape which AI investments the business makes, how it forecasts costs and how it measures returns. Hardy described that expanded role as a partnership with CIOs and other business leaders, each contributing their own expertise.
"Partnering is an important word because I don't think it ever replaces the importance of the other functions," he said.
Managing AI spending
Sixty percent of finance leaders expect AI costs to rise substantially through 2027, according to the trends report. As projects expand, finance teams must forecast how greater usage could affect spending.
Manish Jain, CEO of Strategic Horizon Research, sees the same pressure on finance teams to look beyond approved budgets.
"Traditionally, finance approved tech budgets and tracked spending against them. With AI, CFOs need to forecast how usage will grow, what the service will cost at scale and whether the expected business value will actually show up," Jain said.
So far this calendar year, Salesforce has come to his firm with five new and different pricing models.
John SvioklaChairman and co-founder of GAI Insights
"One of our clients who uses Salesforce reported to us that so far this calendar year, Salesforce has come to his firm with five new and different pricing models," said John Sviokla, chairman and co-founder of GAI Insights.
When spending deviates from forecasts, organizations need controls to identify the cause and respond. The CFO Signals survey suggests those controls remain limited. Fifty-six percent of CFOs reported basic AI-specific controls, such as usage guardrails and team budgets. Another 24% relied on general IT financial controls, while only 20% reported advanced measures such as tracking costs by token or workload.
Without more detailed tracking, IT and finance teams can struggle to identify which AI workloads drive spending increases and where to intervene.
Even when organizations identify an overrun, stopping it can take time. Seventy-four percent of CFOs said their organizations would typically need several business days to several weeks to pause the related AI usage if spending exceeded approved limits.
These gaps require CIOs and finance teams to spot spending spikes sooner and agree on who can limit usage when costs exceed budgets. They must also determine whether growing AI bills reflect useful adoption or spending that fails to deliver enough value.
Deciding which AI investments deserve funding
A rising AI bill can reflect waste, but it can also reflect a useful tool gaining traction. CIOs and CFOs need a way to distinguish the two when deciding whether to expand a project.
Some companies make that decision incrementally. In the trends report, 27% of respondents said they primarily approve large AI and technology investments through staged funding, starting with a pilot and adding money as goals are met. Another 25% require quantified ROI and a formal business case. But 23% said executive or board mandates drive most decisions without a process to measure success.
Staged funding gives leaders opportunities to evaluate results before increasing their commitment. Its usefulness depends on what they ask projects to prove. Expecting an early experiment to demonstrate the same returns as an established deployment can overlook the learning needed to make AI useful, Sviokla argued.
AI is not a technology that can be bought and implemented -- it is a capability that must be grown.
John SvioklaChairman and co-founder of GAI Insights
"AI is not a technology that can be bought and implemented -- it is a capability that must be grown, because people need to interact with this technology to understand it," Sviokla said.
He recommends judging early efforts by whether employees use the tools and develop confidence in applying them. Pilots should then test whether those skills produce business value. Projects ready to scale should face a traditional ROI assessment with clear investment costs, expected outcomes and timelines.
This gives CIOs and CFOs a basis for agreeing on what further funding should accomplish. An experiment might earn support by building useful capabilities, whereas a broader deployment must demonstrate how those capabilities will improve business performance.
The CIO can tell you whether the system works; the business has to show that the work got better; finance should make sure the value reflects in the bottom line.
Manish JainCEO, Strategic Horizon Research
Proving that improvement requires business owners to participate in investment decisions and take responsibility for outcomes. Jain said finance should help establish investment thresholds and measures of value, while CIOs assess technical fit and lead operational controls. The leaders whose teams use AI should be accountable for adoption and results.
"The CIO can tell you whether the system works; the business has to show that the work got better; finance should make sure the value reflects in the bottom line," Jain said.
Tim Murphy is a reporter covering IT strategy for Informa TechTarget, with a focus on IT leadership, governance, workforce skills and AI strategy. His enterprise technology coverage has earned multiple Azbee Awards.
