Every Company Tracks Payroll by Team. Why Not AI Spend?

In the early days of electrification, factory owners had a surprisingly modern problem.

By the late 1800s, businesses across America were replacing steam-powered equipment with electric machinery. Productivity increased, operations became more reliable, and executives knew electricity represented the future. Every month, however, a bill would arrive from the power company with a single number attached to it.

Nobody knew where the electricity was going.

Factory managers could see the total expense, but they couldn't determine which departments consumed the most power or which machines generated the greatest return. They knew electricity made the business better. They just couldn't explain how much better or where the value was being created.

For a while, that wasn't a concern. Electricity was relatively inexpensive, adoption was accelerating, and the benefits seemed obvious enough that nobody spent much time questioning the bill.

As factories expanded, costs increased alongside them. Owners started asking questions that their accounting systems couldn't answer. Why did one facility consume significantly more power than another? Which production lines were actually efficient? Which investments improved output and which simply increased expenses?

The breakthrough wasn't better electricity. It was measurement.

Throughout the early twentieth century, manufacturers adopted submetering systems that tracked energy consumption by building, department, and machine. Electricity transformed from a mysterious overhead expense into something managers could understand and optimize. Once they could see where power was going, they could make better decisions about where to invest.

I've been thinking about that story while watching what is happening with AI.

Most companies can tell you exactly what they spend on engineering. They know what Product Engineering costs. They know what Backend Engineering costs. They know how much each team spends on people and software because they've spent decades building systems to track those expenses.

Now ask similar questions about AI.

Which team generated the most value last month? Which team spent the most? Which tools produced the best outcomes? Which workflows justified their costs?

Many organizations can't answer any of them.

That wasn't a major issue when AI was experimental. A few engineers used ChatGPT, someone purchased a handful of Copilot licenses, and a small innovation budget covered the expenses. Today, AI has become part of the operating budget.

According to Andreessen Horowitz's 2026 Enterprise AI Adoption Report, nearly a third of Fortune 500 companies now have live, paid AI deployments. Coding has emerged as the dominant enterprise AI use case, with organizations spending substantial amounts on Cursor, Claude Code, GitHub Copilot, internal agents, and foundation model APIs.

Yet many engineering leaders still can't tell you what a specific feature cost once AI entered the development process.

Imagine running payroll that way.

Imagine knowing total compensation expenses while having no idea which department employed the people. That's effectively where many organizations are with AI today.

Part of the reason is simple. Companies adopted AI faster than they adopted governance. Engineering teams moved quickly, pilots became production systems, and internal agents multiplied. The focus was on capability because capability was the urgent problem.

The invoice arrived later.

Cloud computing followed a similar path. AWS, Azure, and Google Cloud have supported cost attribution through tagging systems for years, yet many organizations never fully implemented them until spending became significant enough to demand accountability.

AI is now reaching that same moment.

The underlying data already exists. Every token request can be measured. Every API call can be associated with a project, team, user, or environment. The challenge is that the data typically lives with engineering while accountability lives with finance.

Nobody connected the two.

That gap has become large enough that companies such as Stripe, Plaid, and Rippling have reportedly invested in internal tooling to understand where AI spend originates and whether it creates measurable value. When companies operating at that scale begin building the same infrastructure independently, it's usually a signal that the problem is real.

The organizations that solve this first won't necessarily be the ones spending the least on AI. The factories that won the electricity era weren't the ones using the least power. They were the ones that understood how power translated into output.

The same principle applies today. Once you know where the money is going, which teams are creating value, and which workflows deserve additional investment, you can start making better decisions.

And better decisions are usually where the real advantage comes from.