Measure AI Spend as a Labor Cost, Not an IT Line Item

AI spend is another form of a labor cost. It just happens to get billed monthly instead of biweekly.

That's not how most finance organizations are thinking about it today, and I think that's a mistake. My colleague Ben recently made the case for why organizations need to measure the business impact of AI, and I agree with him. But value claims only hold up if you can actually understand the associated cost. 

Before we talk about AI value, we need to talk about what it actually is on the P&L: a labor cost labeled as a software cost.

AI Spend Is a Labor Cost Mislabeled as an IT Cost

AI adds a new flavor of productivity in addition to employees, and it has quickly become a material amount of spend in almost every business. To that end, AI spend should be labeled as a labor cost, not an IT cost.

Employee cost is typically the largest line item on your P&L. It's easy for the CFO to govern and predict those costs as they control employee headcount with the budget. Variations in these costs are controlled through hiring, promotion, and merit budgets.

AI Costs Can Sneak Up on Anyone

One enterprise company CFO shared that their company's AI spend grew to more than 10% of the company’s payroll cost overnight. If that happened at a 50,000-employee company, that's like hiring 5,000 headcount in a month! That is a huge cost that probably wasn’t in their financial forecast.  Now imagine explaining that unfavorable cost variance to your board, or on an earnings call.  

Zylo’s 2026 SaaS Management Index shows how common unexpected AI related cost growth has become. The report found that 78% of IT leaders said they were hit with unexpected charges tied to consumption or AI pricing, and 61% had to cut a project or initiative because of it.

Our own AI spend at Zylo is now twice what we spend on our single most expensive software application. It went from essentially nothing to a material cost in our business. Thankfully, we’ve already built the foundation for successful AI cost management at Zylo and were able to see that cost growth coming. 

Why Companies Are Taken by Surprise

Part of the reason AI costs sneak up on you is visibility—or the lack of it. Most organizations genuinely don't know which AI tools are already used inside their walls. In fact, 60% of IT leaders say they lack full visibility into the AI tools already in use across their own company. 

Lack of visibility stems from shadow AI (shadow IT’s cousin) where individuals and teams purchase AI tools unbeknownst to IT. In 2025, nearly half of the applications in the average company portfolio were expensed by individual employees, and ChatGPT topped that list.

Hidden AI spend is only half the problem. The consumption-based pricing behind it is variable by nature, which makes costs hard to manage and forecast. Costs can swing wildly from week to week, because every employee with a license controls a piece of the spend. It's the equivalent of handing out uncapped corporate credit cards.

Together, hidden AI applications and unpredictable spend are a recipe for financial disaster. You can't govern a cost you can't see, and right now, most CFOs can't see it.

If AI Is a Labor Cost, Measure It Like One

Once you accept that AI is a labor cost, the instinct to just cap spend or chase down every employee with high spend starts to look like the wrong move. 

I've heard this play out at more than one company. Someone flags a high spender as wasteful. They get coached, or in one case, cut off entirely by implementing a spend cap. It later turns out that person was one of the best engineers in the building, generating real value with that spend. Turning their access off didn't save money, but instead nearly cost the company the engineer and the revenue they were driving. 

Spend, on its own, tells you almost nothing about whether the investment is working. That's because spend alone is not the right question. The right question is the one you'd ask about any person on the company’s payroll: what are you getting for it?

AI is labor augmentation or labor avoidance. It belongs in the same equation as the people it's working alongside, not off in its own category. Which means the way to measure it is the same way you'd measure any labor investment: against output.

The Framework: AI-adjusted Cost Per Output

Here's how I think about it. Every part of the business already has metrics it's judged on. Engineering has commits and pull requests metrics. Marketing has pipeline and campaign performance metrics. Whatever the function, there are KPIs already in place for performance management.

Apply AI spend against that same metric, and you'll see a distribution. Some of that spend will move the number. Some of it won't. 

Your true productivity cost is payroll plus AI spend, measured as a ratio against the business result that spend is supposed to produce. That's what turns AI spend into a metric the business and finance can actually manage, the same way we already manage metrics like revenue per FTE or customer acquisition cost.

To come out ahead in this cycle, CFOs will pivot from asking "how much are we spending on AI?" to "what's our AI-adjusted cost per output?" It’s the same question they've already been asking about labor costs for years, now applied to a new category of labor cost — AI.

We've Seen This Movie Before with Cloud Spend & FinOps

Up until 10 or 15 years ago, data center cost management was relatively straightforward to plan and govern. You paid to rent a data center and planned for the capital expenditures to build it out, much like real estate planning. CFOs approved the POs for the hardware and the software then forecasted the capital and depreciation costs. 

Then came AWS with an economic model that made moving infrastructure costs to the cloud a no-brainer. That sounded great on the surface, but it instantly became a nightmare for the CFO from a planning and governance perspective. 

Now the CTO controlled the spend levers, and every decision they made in the cloud had anticipated and unanticipated cost consequences. Leave a job running by accident or spin up multiple experimental environments, and suddenly, your monthly AWS bill blows past the budget before being noticed. 

Those were rough years for finance. The bill was unpredictable, and you didn't know what it looked like until it landed. 

Enter FinOps

In response to this came the creation of FinOps, which is the partnership between Finance and Engineering to ensure infrastructure cloud costs are governed and predictable. 

It requires Finance partnering with the CTO or DevOps team to: 

• Understand the cost drivers.
• Determine how to make costs more predictable. 
• Build the business case with clear ROI
• Agree upon an appropriate level of investment.

Material surprises in spend are not acceptable because they’re budget busters. That's why finance monitors, reports, and re-forecasts every month to prevent surprises.

AI Spend Management Is the Future

Thanks to FinOps, planning, forecasting, and governing cloud costs is no longer the issue it once was. AI, especially with the large language models (LLMs) like Claude and ChatGPT, is just starting down this same path of maturation. 

Over the next several years, I expect AI spend management to evolve into a similar practice, the way FinOps did to optimize, govern, predict and demonstrate ROI for AI spend. 

We're in the early days of AI spend management becoming a practice. The only question is whether you embrace it before it burns you. I'd rather build it now, and not when you are scrambling to fix a material problem in your P&L.

At Zylo, we’ve already built the foundation for successful AI cost management and continue to innovate into this new era of financial spend management. If you have thoughts or questions about managing AI spend, feel free to reach out to me on LinkedIn.

‍