How Finance Teams Should Budget and Govern AI Token Consumption

As generative AI moves from pilot projects to production infrastructure, finance organizations face a new challenge: managing token consumption as a distinct cost category that behaves unlike traditional software or cloud spending. The rapid adoption of AI tools across enterprises has created token bills that can reach triple-digit millions, yet most finance teams lack the visibility and frameworks to govern this spending effectively.

Lukas Deutsch, chief controlling officer at SAP, and David Imbert, chief marketing officer for SAP Financial Management, recently detailed their company's approach to this emerging finance discipline, sharing practical lessons from managing AI costs at scale.

Why it matters

Token consumption represents a fundamentally different cost dynamic than traditional enterprise software. Usage can accelerate unpredictably as employees integrate AI into daily workflows and autonomous agents scale operations. Finance leaders must establish governance that prevents wasteful spending without creating friction that kills productive adoption—a balance that requires new allocation models, ownership structures, and value metrics.

Visibility requires cross-functional collaboration

Most companies initially track AI spending as a single line item, which provides no insight into which teams, applications, or usage patterns drive costs. According to the SAP executives, achieving meaningful visibility demanded collaboration across commercial, engineering, finance, and product functions.

Finance teams contributed forecasting methodologies and accountability frameworks, while operational teams provided context about actual usage patterns and business outcomes. Through repeated forecasting cycles that incorporated more operational detail, SAP's financial models improved substantially. The key lesson: don't wait for perfect data before acting. Start with enough transparency to make informed decisions, then refine the model as consumption patterns become clearer.

Consumption needs owners, not just budgets

SAP's most significant insight was organizational rather than technical. Centralized AI budgets enable early experimentation but disconnect users from financial accountability. As AI embedding deepens across business processes, this model fails.

The company shifted to allocating token costs to specific business areas, transforming conversations from "How much are we spending?" to "What outcomes justify this investment?" and "Should we invest more here?" This approach treats AI consumption like other enterprise resources—software licenses, external services, labor—where decision-makers understand both their consumption and expected returns.

Value metrics matter more than cost per token

Optimizing purely on token cost can destroy value. After deploying AI developer tools, SAP recorded a mid-double-digit percentage increase in pull-request merge rates, indicating meaningfully faster development cycles. The token consumption was justified by measurable productivity gains.

Finance teams need paired metrics: cost data alongside value indicators. Without this dual view, governance risks cutting productive usage simply because the line item is visible and growing.

Surgical controls prevent waste without blocking adoption

As SAP examined consumption patterns, three sources of disproportionate spend emerged: power users and automated agents with concentrated usage, misalignment between task requirements and model selection, and proliferation of redundant tools.

The company implemented targeted controls: token caps to prevent runaway consumption, model routing to match capability and cost to specific tasks, and tool rationalization to eliminate redundancy. These measures contained what the executives described as a triple-digit-million-dollar financial risk while maintaining adoption momentum.

The principle: remove waste while preserving productive demand. Blanket spending restrictions would have been simpler but counterproductive.

Building ongoing governance capabilities

The SAP finance leaders acknowledge their work continues. Next steps include embedding these practices into regular planning cycles, assigning clearer ownership across business units, improving cost allocation accuracy, and building forecasting capabilities that anticipate spending trends before they materialize.

Companies that develop these capabilities won't eliminate AI costs, but they will gain the transparency and accountability to distinguish value-creating consumption from waste—and direct investment accordingly.

These details were first reported by SAP News Center in a feature article by Deutsch and Imbert.