A financial services firm preparing for an IPO discovered, mid-process, that its deal team had been feeding sensitive transaction data into personal ChatGPT accounts for months. No IT approval. No data classification review. No line item in the technology budget. The activity was invisible to every governance function in the organization until an external audit forced it into the open. That firm’s story is not an edge case. It is a preview of what finance leaders will find when they look carefully at how AI is actually being purchased and used inside their organizations.
This is the fourth of five constraints we see between AI investment and earnings. We call it Invisible Spend: the pattern where AI procurement decentralizes faster than finance functions adapt to track it, producing a spending profile that no CFO can consolidate or govern.
The shape of invisible spend
The constraint is structural. AI procurement has decentralized faster than finance functions have adapted to track it. Employees subscribe to tools individually and expense them. Department heads approve SaaS platforms with embedded AI features without flagging them to IT or the CFO. Teams build workflows on top of free tiers of foundation models, then upgrade to paid plans when usage scales, often without a formal purchase order. The result is a spending pattern that is fragmented by design, even when no single actor intends to obscure it.
An analysis by Help Net Security found that 89 percent of AI activity inside organizations goes unseen by IT and security teams. The more conservative figure from Netskope’s 2026 telemetry, that 47 percent of shadow AI usage runs through personal accounts, points the same direction: most enterprise AI spend is invisible to the functions responsible for governing it. When finance leaders actually attempt to produce a consolidated AI spend number, they cannot.
When spend is invisible, ROI is unmeasurable. When ROI is unmeasurable, the board has no basis for deciding where to invest more and where to stop.
What it costs
The cost has two components, and organizations typically underestimate both. The first is direct financial waste. Duplicate tools proliferate. Licenses go unused. Teams in separate business units pay independently for capabilities that already exist elsewhere in the organization. One estimate puts the annual cost of shadow AI incidents, including breaches and wasted licenses, at over $400,000 per enterprise.
The second cost is strategic. Average enterprise AI budgets are projected to reach $85,521 per month in 2025, up 36 percent from the prior year, yet only 51 percent of organizations report that they can confidently evaluate whether their AI investments are delivering returns. Enterprise generative AI investment is projected to grow another 50 percent in the near term. But only 6 percent of companies report achieving 75 percent or more of their expected ROI. The CFO is not managing an AI portfolio. She is managing a collection of unrelated line items that no one has assembled into a coherent picture.
The discipline of spend visibility
Organizations that escape this pattern treat AI spend visibility as a precondition for AI strategy, not a byproduct of it. That means tagging AI resources at the point of purchase, attributing cost by team, project, and workload, and requiring that any AI-related expense, whether a SaaS subscription with embedded AI features or a direct API contract, flow through a defined approval and tracking process. It means regular audits of sanctioned applications to surface the embedded AI capabilities that teams are already using without governance awareness.
Critically, it means building a single consolidated spend register that the CFO can interrogate at any point in the fiscal year. The organizations that do this consistently find that their actual AI spend is materially higher than what finance previously tracked, often 30 to 60 percent higher once shadow usage and embedded tool costs are surfaced. That discovery is uncomfortable. It is also the only honest starting point for converting AI investment into measurable earnings.
The firms that will capture durable value from AI are not necessarily the ones spending the most. They are the ones that know what they are spending, why, and whether it is working.
This is the fourth in a five-part series on the constraints between AI investment and earnings. The two-week Readiness Assessment names which are operating in your organization, verifiable against your own data, before Q3 planning.