AI使用指标: CFOs Struggle to Track Enterprise AI Use

Artificial intelligence has rapidly woven itself into the fabric of nearly every modern company. From sales teams drafting emails with ChatGPT to engineers leveraging GitHub Copilot for code completion, AI tools are ubiquitous. Yet, a significant blind spot is keeping CFOs up at night: they simply can't pinpoint how much AI their employees are actually consuming.

A recent Wall Street Journal piece underscored this critical pain point: AI usage metrics are becoming a major headache for financial executives. Unlike cloud computing, which offers clear API call counts or storage consumption, AI tool usage is fragmented across hundreds of SaaS platforms, internally deployed models, and even personal accounts employees use on the sly.

Why AI Consumption is So Hard to Quantify

A primary reason for this opacity is the coarse granularity of billing. Most AI services are priced per seat or as fixed-tier subscriptions, rather than by granular metrics like tokens or API calls. This means finance departments see a static monthly expenditure without understanding the actual computational resources consumed. To complicate matters further, many teams use personal credit cards for subscriptions like ChatGPT Plus, bypassing corporate procurement systems entirely.

Another significant issue is the proliferation of shadow AI. A developer might pull a model from Hugging Face and run it on an internal server, burying the cost within the broader IT operations budget. Marketing teams generating images with Midjourney often expense these directly from personal accounts. By the time a CFO sees a consolidated report, AI expenditures might have quietly doubled, leaving a trail of untracked spending.

The Ripple Effect on Financial Planning

The inability to track usage directly leads to budget overruns. A finance lead at a SaaS company recently shared that their AI-related spending surged by 300% this year, yet no one could definitively link that expenditure to specific revenue generation. The more profound impact, however, is that ROI calculations become virtually impossible. Without knowing the cost per generation or per inference call, it's incredibly difficult to assess whether an AI project is truly worth continued investment.

This also presents a ticking time bomb for audit and compliance. Certain regulated industries, such as finance, require detailed records of AI usage scenarios (e.g., for email generation). If a company can't even produce a comprehensive inventory of its AI tools, passing an audit becomes a daunting challenge.

Three Pragmatic Recommendations for CFOs

While the industry is still evolving best practices, companies that have already navigated these waters offer some immediately actionable strategies:

• Establish a mandatory AI procurement registry: Require all AI tool purchases, even free tiers registered with personal emails, to be reported. A simple spreadsheet can capture a significant portion of shadow spending.
• Centralize procurement for core AI services: Transition high-volume AI platforms (like enterprise ChatGPT or GitHub Copilot) to usage-based business contracts. This provides finance with detailed billing, where tokens and API call counts serve as excellent unit cost metrics.
• Conduct quarterly 'AI spend audits': Compare AI expenditures against tangible business outcomes, such as customer response rates or code commit volumes. The goal isn't perfect precision initially, but rather to establish a directional understanding of value for money.

Ultimately, CFOs grappling with AI usage tracking today are facing a challenge reminiscent of the early days of cloud computing. Back then, finance struggled to quantify server CPU usage until practices like FinOps emerged. The AI domain will undoubtedly develop its own standardized frameworks for measurement and cost governance. For finance teams, proactively establishing tracking mechanisms now will prevent unwelcome surprises during next year's budget season.