Neil Sawyer, Senior Vice President and Managing Director for HP Europe, Middle East and Africa, recently published “Scaling AI Without Losing Control” in Forbes Business Council.
His starting point will be familiar to any finance leader. AI pilots looked affordable on pay-per-token and pay-per-query pricing, and that same pricing can get far more expensive at enterprise scale. Gartner forecasts worldwide AI spending will reach $2.7 trillion in 2026, up 49.5%.
He organizes around three questions:
- What are we spending?
- What value are we creating?
- Where should each AI task run?
He then turns to governance, and says finance and IT should share responsibility for it.
He’s asking the right questions, and our own research shows how far most finance teams are from answering them. In CloudZero’s 2026 State of AI ROI report, a survey of 252 CFOs and VPs of Finance, 87% said they are increasing FY2027 AI budgets, even though 61% already spend more on AI than they can justify with measurable outcomes. Allocation is the weak point, with 54% able to see AI spend only as a single bill line or a rough category split. Closing that gap takes cost data by team, feature, customer, and agent, which is the problem CloudZero was built to solve.
Here is how CloudZero answers each question in practice.
What are we spending?
Before a CFO can control AI spend, finance needs to know the total, plus spend by team and by tool.
Providers bill AI as consumption charges, and none of the invoices tell finance which team, feature, or customer drove them.
CloudZero uses dimensional allocation to assign that spend to teams, features, and customers, so finance knows who owns each part of the total and where tools overlap.
2026 State of AI Spend Report
Why Finance Caps What It Can’t See
Learn how finance leaders are managing AI spend, and how they’re approaching FY2027
What value are we creating?
High token usage doesn’t prove business value. KPMG’s Q3 2026 Global AI Pulse found that 89% of organizations report measurable value from AI, but only 12% consistently assess that value against what it costs. Leaders need to measure outcomes such as hours saved or higher conversion rates.
AI ROI needs both sides measured. The business supplies the outcome. Finance needs the cost side at the same level of detail, such as cost per customer, per feature, per agent, or per resolution.
With CloudZero AI Signals, finance can trace each AI usage event to the team, feature, customer, and activity behind it. The figures reconcile to the invoice, so finance can defend them.
Where should each AI task run?
The third question is about AI workload placement, which means choosing between a large cloud-hosted AI model and local processing.
CloudZero supplies the cost per AI task that the placement decision depends on. A team can’t judge whether an AI workload belongs on a large cloud model without knowing what the task costs today, and that is a unit economics question.
The same cost question applies among cloud AI models, where a team can end up paying for a larger model than the task requires. Few organizations make that choice deliberately. KPMG found model routing, the capability that decides which model handles which task, was the least common piece of enterprise AI management, in place at 23% of organizations.
CloudZero’s Model Rightsizer helps teams match each AI task to the smallest model that does the job. It’s a free, open-source plugin for Claude Code. It scores each task on how much quality would suffer if a smaller model handled it, and on how much cost, speed, and volume pressure the task carries. For each task, it recommends a model with a confidence level, a runner-up, and the condition that would change the pick, plus a reasoning effort and a token budget. Afterward, a usage report compares actual AI spend from CloudZero AI telemetry with the plan, so finance can see the difference.
In one week of one CloudZero team’s work in Claude Code, spend on Claude Opus fell about 75%, from $2,813 to $711. That’s one team over one week, so test it on your own AI workloads and check output quality before you set policy.
Governance needs numbers behind it
Good governance means clear policies on approved tools, ownership, and budgets, plus regular reviews of whether the spend delivers outcomes. Teams can only enforce those policies with data behind them. To assign ownership, finance needs every AI dollar mapped to a team.
Shared responsibility matters because the office that owns the AI budget usually can’t see how it’s being spent. In CloudZero’s survey, 65% of CFO-led AI programs fell in the two weakest allocation tiers. To enforce a budget, teams need to know about a spike while it is happening, before the invoice arrives.
Timing is the other gap. CloudZero found 26% of finance leaders said decision-makers get AI spend data only at the end of the billing period or later, after decisions are already made.
Decide what to scale next
Deploying AI with discipline takes cost data at the level where decisions get made, which means by team, by feature, by customer, and by agent. With that data, a CFO can explain AI spend, justify it, and decide what to scale next.
If your team is working through those three questions right now, start with the first one. Book a demo to see how CloudZero ties your AI spend to the teams, features, and customers behind it.
