AI Spending Visibility: Only 1 in 4 Companies Knows What It Pays
Agentic AIThree-quarters of companies cannot fully account for what they’re spending on AI — and as agentic workloads push token usage into autonomous, multi-hour sessions, that blindspot is about to become a finance department crisis.
The Number
A KPMG survey reported by The Decoder found that only 26% of companies have full visibility into their AI costs. Another 50% have limited oversight, and 22% have no transparency at all — discovering expenses only when the billing statement arrives.
The gap between adoption speed and financial governance is already generating real consequences. One KPMG client saw a sixfold increase in token usage after deploying agentic workflows. Several others exhausted their entire annual token and cloud budgets within months of initial deployment.
How the Math Works
Token-based billing behaves differently from every cost model that preceded it. SaaS subscriptions are predictable by seat count. Infrastructure is capacity-constrained and scales with provisioning decisions. Tokens scale with behavior — and in an agentic context, behavior is difficult to predict or bound.
A developer prompt that initiates an autonomous coding session can consume more tokens than weeks of standard conversational use. Steve Chase, KPMG’s global AI lead, describes it as “a new resource that needs to be managed that didn’t exist quite that way,” adding that the industry is “seeing exponential growth” in these costs.
The billing pressure point is arriving fast. Gil Luria, head of technology research at D.A. Davidson, predicted: “A lot of CFOs are going to see their Anthropic bill and freak out this quarter.” The pattern echoes the cloud infrastructure overspend that followed the pandemic — companies adopted rapidly, financial tracking followed slowly, and cost cuts came after the shock.
What It Signals
This is the maturity gap in enterprise AI adoption. Technical teams have moved to agentic workloads; the financial governance layer has not moved with them. Token spend is not inherently wasteful — but untracked spending produces both budget overruns and perverse incentives. Employees at Meta and Amazon, according to the same reporting, gamed internal AI productivity leaderboards by consuming tokens on purposeless tasks rather than value-generating ones.
Finance teams that managed software costs through seat counts and annual renewals are unprepared for per-call, per-token, per-model billing that compounds in real time across every team that has provisioned a model API key. Unlike B2B AI adoption data from Ramp — which shows which vendors are winning — visibility data shows whether companies can even see what they’re buying.
The parallel to cloud sprawl is worth taking seriously. Cloud overspend became a structural problem because adoption preceded governance tooling by two to three years. AI token spend is moving on the same curve, but with faster adoption rates and more complex billing surfaces.
The So What
The companies that will extract disproportionate value from agentic AI are the ones that treat token governance as a first-class engineering concern — not an accounting artifact. That means per-project token budgets, agent task scoping that defines abort conditions before a session starts, and visibility dashboards that make spending legible before it becomes a surprise. The blindspot identified in this data is not just a CFO problem: it is a signal that most agentic AI deployments still lack the discipline that production infrastructure requires.
Source: The Decoder
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →