
AI budget ownership shifts to the CFO as spend exits IT
With roughly 80% of AI spending now originating outside the IT function, the 2027 plan forces finance chiefs to settle approval thresholds, chargeback rules and what an autonomous agent is allowed to spend.
AI budget ownership has quietly become a CFO problem rather than a CIO problem. TechTarget reported in October that roughly 80% of AI spending now originates outside the IT function, spread across marketing, customer service, engineering and finance itself. That single statistic undoes the assumption behind a decade of technology governance: that the IT budget line is where technology spend gets authorized, measured and stopped.
Why the IT budget line stopped being the control point
For most of the cloud era, the architecture of control was simple. Software purchases ran through IT, procurement enforced the contract template, and the CIO's budget envelope acted as a natural ceiling. A marketing team could want a tool, but it could not provision one at scale without a signature from someone whose cost center was visible to finance.
Generative AI broke that chain in two ways. Capability is now embedded inside applications business units already own, so adoption does not require a new vendor relationship. And pricing has detached from headcount. Token draws, per-resolution fees and agent runtime scale with usage, which means consumption grows without anyone making a second purchasing decision. Bloomberg's CFO newsletter framed token cost as the item keeping finance chiefs awake, and the reason is structural rather than psychological: variable cost with no approval event is the hardest thing on a forecast.
The result is that the CFO inherits consumption that was never approved in a capital cycle. Survey coverage in CFO.com found 32% of organizations exceeded their AI budgets by at least 10% in the 12 months before July 2026 fieldwork. Overruns of that size are not a planning error. They are the signature of a spend category with no gate.
Pressure to deploy is colliding with thin ROI evidence
The pressure side of the equation is well documented. The same CFO.com survey work put 76% of finance leaders under high pressure to fully implement AI. Deloitte's 4Q25 CFO Signals found 87% of CFOs expect AI to be very or extremely important to finance operations in the year ahead. Very few of those same leaders can point to settled return evidence.
That gap matters for how the budget conversation runs. When deployment pressure is high and measured return is thin, the default behavior is to fund experiments broadly and let the business units sort out value. That is a defensible posture for one cycle. It is not defensible across two, because the spend becomes a run-rate before anyone has decided which use cases earned it.
Vendor pricing is compounding the problem. Deloitte's June 2026 Technology Spotlight on outcome-based pricing documented agentic software vendors billing per result rather than per seat. Per-result pricing aligns incentives neatly in theory. In practice it makes spend genuinely variable and close to impossible to pre-approve in a traditional annual plan, because volume is a function of business activity rather than license count.
Three decisions that land before the 2027 plan closes
The first is the approval threshold for consumption-based contracts that carry no ceiling. Most delegation-of-authority matrices are written against contract value. A consumption agreement has no contract value at signature, which means it often clears authority tables designed for fixed commitments. Finance teams are rewriting thresholds to trigger on projected annualized consumption, with hard caps and automatic review at defined burn levels.
The second is chargeback versus central absorption. If business-unit AI spend is absorbed centrally, segment margins look artificially healthy and the units face no price signal. If it is charged back, unit economics by segment change materially and some product lines will look worse than the board expects. Neither answer is wrong, but the choice has to be made deliberately rather than inherited from how cloud costs happened to be allocated in 2019.
The third is spend authority for autonomous agents. PYMNTS reported in October that CFOs are building approval controls finance agents cannot override, with an emerging rule that no single actor should be able to request, approve and execute a movement of money. That principle is familiar from segregation of duties. Applying it to a software agent means deciding, explicitly, which human signs for what the agent initiates.
Regulators are arriving before the controls are finished
The UK Financial Conduct Authority signaled in October that it is drafting a framework for agentic finance, citing operational failure and cyber exposure as the driving concerns. That is an early marker, not a rule, but it tells finance leaders something useful about sequencing. Supervisors are going to ask who authorized an agent to act, what limits applied, and what the audit trail shows.
Firms that have already answered those questions for internal budget reasons will find the compliance work largely done. Firms that have not will be reconstructing decision rights under external deadline pressure, which is the expensive version.
On the supply side, the build continues. The AWS CFO argued in an October Axios interview that the infrastructure expansion is sustainable. Whether or not that holds, it sharpens the buyer-side question. Capacity will be available. The constraint is not supply. It is whether the buyer has decided who is allowed to consume it.
What a defensible position looks like this quarter
A workable position has four components. A consolidated view of AI spend across all cost centers, not just IT, so the total is known before it is debated. Authority thresholds written against projected consumption rather than contract value. A stated allocation policy, chargeback or central, applied consistently. And a documented control rule for agent-initiated transactions that preserves separation between request, approval and execution.
None of that requires resolving the ROI question, which will take longer than this planning cycle. It requires accepting that AI budget ownership has already moved to finance by default, and deciding to exercise it on purpose. Teams that skip the step will not avoid the decision. They will make it in a Q2 reforecast, with less room and worse information.
Key takeaways
- Roughly 80% of AI spending now originates outside IT, so the IT budget no longer functions as a control point for technology consumption.
- 32% of surveyed organizations exceeded AI budgets by at least 10%, a signature of variable spend with no approval gate.
- Rewrite delegation-of-authority thresholds to trigger on projected annualized consumption, not contract value at signature.
- Decide deliberately between chargeback and central absorption, because the choice changes segment unit economics.
- Apply segregation of duties to agents: no single actor should request, approve and execute a movement of money.


