The CFO’s guide to governing runaway artificial intelligence spend

Token-based pricing has made artificial intelligence (AI) one of the hardest costs in the enterprise to forecast, and finance now owns the response.

Key takeaways 

• Consumption-based AI pricing scales with usage, not budget, so a single model upgrade or high-volume workflow can multiply costs in weeks.

• Most finance teams can’t trace token spend to specific teams, products, or use cases, leaving ROI hard to prove and accountability hard to enforce.

• CFOs who model cost by workflow, connect spend to outcomes, and govern adoption from the start turn unpredictable AI bills into a managed, accountable program.

The invoice arrives. The number doesn’t match the projection, and the CFO is the last to understand why.

This scenario plays out across enterprises of all sizes as AI adoption moves faster than the financial controls set up to contain it. What started as a controlled experiment has quietly grown into a sprawling, consumption-driven cost line that most finance teams aren’t able to track. AI spend is now unpredictable, often invisible, and growing faster than any budget model expected, making it a finance problem that demands a CFO response.

Why token-based pricing breaks the budget model

Traditional enterprise software costs were predictable. You paid per seat, subscription, or license, finance knew what to expect, and budget variances stayed manageable.

AI pricing works differently. Many providers now charge based on the volume of tokens processed rather than a flat fee. Cost scales with how much AI teams actually use, not what they contracted for. That single shift breaks the assumptions behind most enterprise budget models. Unit prices have fallen, yet enterprise AI invoices keep climbing as usage growth outpaces savings.

Agentic AI systems, the kind that complete multi-step tasks on their own, consume far more tokens per task than standard conversational tools. Every loop, validation, and context refresh adds to the bill, and the way those workloads are routed through models can swing costs dramatically—often without oversight from your finance function.

Why finance can’t trace what’s being spent

The function that spent a decade building governance for cloud infrastructure has inherited a very different cost structure driven by token volume, retrieval overhead, and agent loops.

The cost drivers hiding behind those invoices rarely show up in a vendor proposal. Retrieval overhead, the work of pulling relevant documents into context before a query runs, can inflate input costs. Agent loop retries, where a system resubmits a task with the full conversation history when output falls short, multiply token use quickly. Background inference from monitoring and compliance systems runs continuously, with no easy way to throttle it.

None of this appears on a dashboard or was modeled before deployment. And most finance teams lack the tooling, vocabulary, and workflow visibility to catch it before the invoice lands. This contributes to AI “sprawl,” when usage and spend become hard to track and align to concrete goals.

The ROI measurement gap finance can’t ignore

Rising costs are easier to justify when the returns are clear. Many AI initiatives fail to deliver the return leaders expect, and much of that comes down to measurement. Most organizations approve AI investments without setting baseline metrics, attribution frameworks, or time-to-value milestones, sending “sprawl” into overdrive as it starts to affect several different areas across your organization. When the ROI question arrives, finance has no clean denominator to work from.

Overspending is only part of the exposure. Finance also risks mistaking activity for impact, renewing tools on the strength of perceived innovation rather than verified outcomes, and compounding that misallocation across every budget cycle.

Balancing cost discipline with competitive pressure

AI-related purchases often clear approval faster than any other software category, and teams sometimes tie a purchase to an AI initiative specifically to speed it through. At the same time, many of those same organizations are told to cut costs or reduce vendors.

Finance is being asked to tighten budgets and speed up AI adoption at once, with no shared framework for deciding which investments clear the bar. The result is a budgeting loophole. AI became a label that skipped scrutiny and projects stamped with AI cleared approval faster than the oversight to evaluate them.

The CFO’s role is expanding in response. Finance is increasingly asked to lead AI governance and act as a strategy architect, connecting finance, technology, tax, and HR, and most teams aren’t yet equipped for the role.

The risk of decentralized adoption and shadow spend

Governance gaps widen when AI purchasing spreads across the organization. Teams buy tools without procurement involvement, pilots skip formal review, and few organizations have a mature AI governance program in place.

The result is AI “sprawl”: unofficial tools, free-tier apps, and browser extensions running outside IT’s frameworks, storing data in places neither finance nor IT monitors. When an approved tool doesn’t fit how people work, employees reach for whatever gets the job done. That fragments data, muddies reporting, and creates cost exposure that stays hidden until something breaks.

“Every AI decision you make has to come from a place of readiness,” says Amir Shah, Strategy and Transformation Partner at Highspring. “Everyone talks about clean data foundations. But cleaning data without accounting for process and then unleashing AI is a disaster waiting to happen and creates years of workarounds and pushes spend to unsustainable levels.”

For regulated industries, including financial services, healthcare, aerospace, and defense, this exposure carries real weight. Governance failures in these sectors bring audit, regulatory, and reputational risk that can dwarf the AI spend itself. Organizations can maintain audit-preparedness, maximize ROI, and create lasting value with AI by aligning data, governance, and target operational models with enterprise platforms like NetSuite and OneStream.

How Highspring helps CFOs bring discipline to AI investment

Turning unpredictable AI spend into a governed, accountable program calls for expertise at the intersection of finance, technology, and strategy.

The organizations that manage AI spend effectively share a consistent approach: modeling token volume by workflow type before finalizing architecture, building governance that connects spend, usage, risk, and ROI in a single view, and making finance an active participant in AI decisions from the start. The CFOs who lead well will be the ones who build the governance that lets AI move faster with fewer surprises: clear ownership of ROI tracking, token-level visibility by workflow and team, architecture decisions made with cost models attached, and reinvestment strategies that turn efficiency gains into measurable outcomes.

Highspring helps organizations build the measurement infrastructure, governance frameworks, and financial controls that AI investment requires. That means setting clear ROI baselines before deployment, building cross-functional accountability across finance, operations, and data teams, and designing reinvestment strategies for the productivity gains AI creates, so those gains compound rather than stall.