Generative AI has moved from a line item in the IT budget to a daily operational expense for many companies, and finance leaders are now tasked with managing a resource most were never trained to forecast: AI token consumption. At SAP, the internal effort to build a framework for this spending has surfaced a set of practices that shift the conversation from cost control to value governance.
Start with visibility, not precision
Most organizations begin by tracking AI spend as a single figure. That number tells you the total bill but nothing about which teams, workloads, or usage patterns are driving it, or whether the consumption produces useful outcomes. Lukas Deutsch, SAP's chief controlling officer, and David Imbert, chief marketing officer for SAP Financial Management, said that getting better visibility required collaboration across commercial, engineering, finance, and product teams. Finance brought the forecasting questions and accountability structure. Other functions supplied the operational context that made the numbers meaningful.
Repeated forecasting cycles improved SAP's financial models as more operational detail was layered in. The lesson was not to wait for perfect data before acting. Start with enough transparency to make better decisions, then sharpen the model as patterns emerge.
Assign ownership or lose accountability
A centralized AI budget makes early experimentation easy, but it also disconnects the people spending tokens from any financial responsibility for them. As AI embeds deeper into business processes, that model breaks. "Our most consequential insight was philosophical rather than financial," Deutsch and Imbert said. "We learned that visibility alone isn't enough and that consumption needs an owner."
This does not mean charging back every large language model call with forensic accuracy. It means managing AI consumption the same way companies manage software, external services, and labor. At SAP, allocating token costs to business areas shifted the internal discussion from "How much are we spending?" to "What are we getting for this?" and "Is this the right place to invest more?" Token spend becomes a strategic and operational investment decision rather than a technology line item.
Cost per token is the wrong scorecard
A large AI bill draws attention, but optimizing purely on cost can destroy more value than it saves. After SAP rolled out AI developer tools, the company recorded a mid-double-digit percentage increase in pull-request merge rates, a clear signal that development work was accelerating. The consumption was worth it. Finance needs a paired view: cost metrics alongside value metrics. Governance without that view risks cutting usage that is actually generating returns.
For finance professionals building their own frameworks, resources like the AI Learning Path for CFOs can provide a structure for connecting AI investment to business outcomes. The core principle is that an AI tool that meaningfully accelerates development, reduces repetitive work, or improves customer service will carry real token costs, and that is acceptable when the value is measured.
Target waste, not adoption
As usage scales, controls become necessary, but the right controls are surgical. When SAP examined consumption patterns, three root causes of disproportionate spend emerged: power-user and automated-agent concentration, model misalignment, and tool proliferation. The response centered on three levers: token capping to prevent runaway consumption, model routing to match capability and cost to the task, and tool rationalization to eliminate redundancy. That work helped contain a triple-digit-million-dollar financial risk while keeping adoption moving forward.
"Guardrails should target waste, not adoption," Deutsch and Imbert said. The principle is to remove waste while preserving productive demand. This approach also surfaces insights relevant to broader AI for Finance strategies, where the same discipline of pairing cost data with operational context applies.
Why this matters for finance professionals
AI token spend is becoming a recurring enterprise cost that will appear in budgets with increasing frequency. Finance teams that treat it as a single IT line item will lack the visibility to distinguish productive consumption from waste. The alternative is to build a governance model that assigns ownership, pairs cost metrics with value metrics, and applies controls that target the sources of avoidable spend rather than putting a blanket brake on usage. Companies that do this will still have an AI bill to pay, but they will have the transparency to direct investment toward what actually creates value.
Your membership also unlocks:
