AI Doesn't Just Need Guardrails. It Needs Receipts. How Governance-by-Design Gets You There

Finance chiefs are trying to get a better read on how much autonomy their companies are buying when they “buy AI,” as vendors increasingly charge by usage and business teams increasingly deploy agents that don’t just draft content—they take actions.

The shift to token-based pricing is colliding with a second shift that’s harder to see in a budget spreadsheet: the push toward autonomous AI, where systems observe, decide, and act inside workflows with less human supervision.

That combination is creating a new kind of operational risk for CFOs and risk leaders alike: costs that can jump suddenly, and decisions that can’t be reconstructed quickly when someone asks, “What happened, and why?”

The Wall Street Journal’s CFO Journal recently put numbers on the visibility problem. Only 26% of companies say they have a comprehensive view of their AI costs, while 50% have some visibility and 22% report no visibility or visibility after billing, based on an as‑yet‑unreleased KPMG survey. KPMG’s Steve Chase told the Journal that token usage is growing “exponentially,” with some companies blowing through annual token and cloud budgets in months and one client’s token usage rising sixfold.

Gartner, meanwhile, is telling executives to plan for more autonomy, not less. In its Business Quarterly (2Q26), Gartner argues that “autonomous business is how AI will be monetized,” and says the shift “will gain pace over the next five to 15 years.” It also warns leaders not to treat autonomy as a future optionality: eight in 10 executives think autonomous business will be the predominant form of business by 2030, and 76% of CEOs view AI as the No. 1 technology likely to disrupt their industries over the next three years. Gartner adds an uncomfortable readiness signal: only 44% of chief executives consider their own CIOs AI savvy.

Put those threads together, and you get a tension that’s starting to show up in budget meetings: the board wants AI-driven growth, the business wants agents in production, vendors want to meter usage, and finance wants predictability. The missing ingredient is often not model performance. It’s the information foundation that makes autonomous AI governable—and therefore financially manageable.

That’s where Information Governance (IG) by Design enters as something closer to a control system than a compliance slogan. IG by Design starts from a simple premise: if you can’t reliably govern the information surrounding AI—inputs, configurations, versions, access, and evidence trails—you can’t reliably govern outcomes. And when the pricing unit is tokens, you also can’t reliably govern spend.

The evidentiary questions described in automated hiring illustrate the point. When an applicant is rejected by an opaque scoring system, the hard questions are operational: what data did the system use at decision time, which version of the scoring logic ran, were knockout questions configured (and by whom), what threshold was applied, and what changed since last quarter? When those answers are scattered across vendor logs, HR systems, and ad hoc spreadsheets, an organization can’t prove governance even if it believes it behaved responsibly.

Token pricing makes those same questions suddenly financial. When the CFO asks why AI spend doubled, “usage went up” isn’t a usable explanation. The usable explanation is a reconstruction: which workflow drove the increase, which agent version ran, which model and context window were used, how many retries occurred, which tools were called, what configuration changed, who approved it, and what business outcome it produced. Without that chain of evidence, cost containment tends to devolve into blunt restrictions—freezing experimentation, capping usage indiscriminately, and slowing the productivity gains that justified AI investment.

Gartner’s recommended success metrics for autonomous business quietly reinforce this linkage between governance and cost. It advises companies to avoid “death by a thousand use cases” and measure outcomes across value and operational autonomy, including revenue per employee and new revenue streams from autonomous sources, but also the “shift from human-operated to machine-operated” using metrics such as human-in-the-loop reduction and human-tech ratio (tracking IT spend and human FTE spend as shares of operating expenses). For agentic AI specifically, Gartner calls out token consumption and latency as technical performance measures to gauge whether an autonomous solution is scalable and cost-effective. In other words, “tokens” aren’t just a billing artifact; they’re now a success metric—and that makes governance design inseparable from financial control.

Companies are also learning that “human in the loop” is not automatically a control. If a reviewer only sees a ranked list and clicks approve, that’s not oversight—it’s delegation with a signature. HITL becomes a governance mechanism only when it is auditable: the reviewer’s role is defined, intervention points are explicit, the rubric is documented, and the system captures what the human saw, what they changed, and why. In a token-priced environment, that same design discipline forces clarity about where autonomy is allowed to run freely and where it must slow down, justify itself, or escalate—preventing runaway agent loops and “invisible” compute spend from becoming a recurring surprise.

The autonomous business narrative says competitive advantage will come from increasing machine agency across operations, workforce workflows, and products. The CFO narrative says the meter is running, and it’s harder to forecast than seat-based SaaS ever was. IG by Design is the bridge: it turns autonomous AI from a black-box expense into a governable system with receipts.

The organizations that scale autonomous AI successfully are unlikely to be the ones that “use the most tokens.” They’ll be the ones that can tie usage to outcomes, and outcomes to evidence—showing, at any moment, what the system did, why it did it, who was accountable, what it cost, and whether it paid off.