In the early twentieth century, manufacturing underwent a quiet revolution. Factory owners had always known how much money they spent running their operations. They knew how much raw material arrived at the loading dock and how much finished product left the building. They could see profits rise and fall on financial statements. What they struggled to understand was where performance actually came from.
When production slowed, managers often blamed the wrong department. When costs increased, nobody could confidently explain why. The information existed somewhere inside the organization, but it was scattered across different people, processes, and machines. A factory could be operating inefficiently for months before anyone identified the source of the problem.
The solution wasn't better accounting. It was attribution.
As manufacturers began measuring output by production line, tracking performance by department, and connecting costs to specific activities, something interesting happened. Problems that once felt mysterious became obvious. Leaders no longer had to speculate about what was happening inside the business because they could finally see it.
A surprising number of companies are discovering they have the same problem with AI.
The monthly OpenAI invoice arrives, and the total is significantly higher than it was thirty days ago. Finance notices immediately. Engineering gets a message asking for an explanation. A handful of theories emerge. Maybe a new feature launched. Maybe usage increased. Maybe a team switched models. Maybe an internal project became more popular than expected.
The challenge is that most organizations don't actually know which explanation is correct.
An OpenAI invoice is excellent at telling you how much was spent. It is far less effective at explaining who spent it, which project generated the usage, or whether the increase was connected to a successful initiative or an expensive mistake. By the time activity from product teams, internal tools, support workflows, and experimental projects reaches the invoice, it has usually been compressed into a single number.
That number is useful for accounting purposes, but it doesn't provide much insight into how the business is operating.
Imagine two teams. One launches a customer-facing feature that generates new revenue and dramatically increases API usage. Another builds an internal workflow that quietly burns tokens all day while creating very little value. If both activities flow through the same credentials and infrastructure, they become indistinguishable on the invoice. Leadership can see the cost, but they can't see the story behind the cost.
This is why attribution matters.
The organizations getting the most value from AI aren't necessarily spending less than everyone else. In many cases, they're spending more. The difference is that they understand where the spending occurs, which teams are responsible for it, and what outcomes it produces. They know whether increased usage represents waste, growth, experimentation, or competitive advantage.
That level of understanding starts with a relatively simple decision. Teams need their own credentials. Projects need their own credentials. Internal agents and customer-facing applications need identities that allow their activity to be measured independently. Once that foundation exists, questions that previously required guesswork become easy to answer.
Which team is generating the most AI spend? Which projects are producing the highest returns? Which workflows deserve additional investment? Which initiatives are consuming resources without delivering meaningful results?
The OpenAI invoice won't answer those questions on its own.
It will tell you that something changed.
Attribution tells you what changed, who changed it, and whether you should be happy about it.