Who Owns Your AI Bill? Why Token Spend Needs a Name Next to It

Key Takeaways

• Half of companies can't name who owns AI spend. In a 2026 survey of 700 engineering leaders, 52% said there is no clear, dedicated owner of AI costs in their organization.
• Split accountability is the same as none. When four functions each hold a quarter of the responsibility, every overrun becomes someone else's problem.
• The people who drive the spend rarely see it. Engineers pick the models, write the prompts and set the retry logic, and fewer than half of them know what their work costs to run.
• Attribution comes before accountability. You can't hold a team to a number you can't split by team, project and workflow.
• A name should own value as well as cost. The goal is a person who can say what the spend bought, and defend it when it's worth it.

Every other line on your P&L has a person attached to it. Payroll belongs to someone. Cloud hosting belongs to someone. The office lease belongs to someone, and if it jumps 40% next quarter, everyone knows whose phone rings first.

AI spend is the odd one out. At a lot of companies it lives on a corporate card, a few API keys, a pile of Claude seats and an invoice that finance pays without being able to explain. Harness's 2026 State of AI in FinOps report put a number on it: 52% of the 700 engineering leaders and practitioners it surveyed say their organization has no clear, dedicated owner of AI costs. The same survey found 72% hit a surprise AI bill in the past year. (Harness sells cost management software, so it has a reason to find this problem. I still think the number is about right.)

Those two stats belong together. A surprise bill is what you get when nobody's job is to see it coming.

And this is no longer a side budget. The FinOps Foundation's 2026 State of FinOps survey found 98% of its 1,192 respondents now manage AI spend, up from 31% two years earlier. The spend went mainstream. The ownership didn't follow it.

So the way I've started framing it for the teams we work with is simple. Before you ask how to cut your AI bill, ask whose name is on it. If the answer is "everyone's," the real answer is nobody's.

AI Spend Is the One Line Item Nobody Signs For

Most companies didn't decide to leave AI spend unowned. It happened by accident.

The first AI costs were small. A few seats for the curious people. An API key for an engineer's prototype. A pilot that someone's VP approved out of an innovation budget. None of it was big enough to need an owner, so nobody assigned one.

Then usage grew. Claude Code rolled out to engineering. Ops built a few agents. Sales got seats. Each step was a reasonable call by a reasonable person, and each one added to a bill that still had no name next to it. Anthropic's own guidance on managing Claude Code costs puts enterprise usage at around $13 per developer per active day, or $150 to $250 per developer per month. Multiply that by an engineering org and it stops being a rounding error.

Here's the uncomfortable part. A line item without an owner never gets defended, either. When the CFO looks for something to cut, the unowned line is the easiest target, because nobody shows up to explain what it bought.

Ask who gets the call

Try this test. Imagine your AI bill doubles next month. Who gets the first call from finance? If you can name one person, you have an owner. If the honest answer is "probably a Slack thread," you don't.

Shared Ownership Turns Into No Ownership

The obvious fix is to give AI costs to a team. The problem is that most companies already did, to four of them at once.

The Harness data shows how scattered it is. Asked who is accountable for managing AI costs, respondents split almost evenly: platform and DevOps teams at 30%, dedicated FinOps at 27%, finance at 23% and engineering at 19%. No function came close to a majority.

I get why. AI spend touches all four. Platform owns the infrastructure, FinOps owns the reporting, finance owns the budget and engineering makes the calls that drive usage. Each has a fair claim to a piece of it.

But a cost with four partial owners behaves like a cost with zero. Finance sees the overrun and asks engineering. Engineering says the platform team set the limits. Platform says FinOps owns the reporting. FinOps says it can only report what it can see. Everyone is right, and the bill keeps growing.

Name one owner per budget

Assign ownership the way you would for any other budget. One person per budget line, usually the leader of the business unit or team that spends it. Platform, FinOps and finance still play their parts, but they support the owner instead of each holding a slice. One name. Everyone else helps.

The People Who Spend the Tokens Rarely See the Bill

The deeper problem is where decisions happen versus where costs show up.

Engineers choose the model for a workflow. They decide how much context an agent carries into each call, and how many times it retries when something fails. Those choices drive most of the bill. Yet Harness found that on average only 45% of engineers understand what the things they build cost to run, and 56% of engineering leaders say their teams don't have cost in mind when building AI features.

Engineers care about cost. The data just isn't in front of them when they make the call. They see a model that works well. Finance sees an invoice a month later. Those two views never meet.

Meanwhile, plenty of companies are actively pushing usage up. The same survey found 57% of organizations encourage what it calls "tokenmaxxing," pushing people to use as much AI as possible regardless of what it returns. I understand the instinct. You wanted adoption, so you rewarded usage. But usage targets with no owner and no cost signal are a recipe for exactly the surprise bills the survey describes.

Put the cost where the decision gets made

The fix is to move cost data closer to the people making the choices. A team lead should see their team's spend next to their team's output, weekly. An engineer building an agent should know roughly what one run costs before it ships. Claude's own tooling now supports some of this: Team and Enterprise admins get a spend report by user and model and can set spend limits by organization, group or member. Use those controls, and give the numbers to the people who can actually change them.

Attribution Comes Before Accountability

You can't hold someone accountable for a number you can't split.

This is where a lot of ownership efforts stall. Leadership names an owner, and the owner opens the provider dashboard and finds one total by day. No team breakdown. No project. No workflow. They now own a number they can't explain, which is worse than not owning it.

Harness asked a sharp question here: if your AI spend doubled overnight, how long would it take to identify the source? Only 20% said within hours. 40% said a full day, 32% said a week, and 8% said longer or never. A week of not knowing is a week of paying for it.

The usual reason is plumbing. Usage was never tagged at the source, so there's nothing to slice later. Seats were handed out without mapping them to teams. API keys were shared across projects. By the time someone needs the history, it doesn't exist.

Tag spend at the source

Before you name an owner, make sure they'll have something to own. Map every seat to a team. Give every project or agent its own key or workspace. Agree on what a "workflow" is in your company so spend can roll up to it. This is boring setup work, and it is the work that makes every later conversation possible.

Every AI Bill Needs Four Names, and One of Them Signs

Here's the structure I'd suggest. Everyone has a role, and one person signs.

The budget owner is the name next to the number. The other three exist so that person can actually do the job. Without a platform owner, there's no attribution. Without a reviewer, nobody notices the drift until the quarterly close. Without workflow owners, the budget owner has to understand every agent personally, which doesn't scale past a handful of them.

Write the owners down

Put this table in a shared doc with real names in it. Review it whenever a new team or agent goes live. A new workflow without a named owner is how unowned spend starts all over again.

An Owner Defends Value Before Cutting Cost

There's a failure mode on the other side, too. Give someone ownership of a cost, and their first instinct is to shrink it. I'd push back on that instinct.

The point of an owner is to be able to say what the money bought. Some teams should be spending more. An engineering team that ships twice as fast with Claude Code is a bargain at almost any reasonable spend. An agent that costs a couple of dollars a run and saves someone an hour is worth funding. The owner should be the person who makes that case, with numbers, when someone upstairs wants a blanket cut.

That's harder than it sounds today. Harness found only 26% of organizations have a solid method for measuring the business value of their AI spend. Respondents also estimated that 26% of their AI spend is wasted. Those two numbers are related. If you can't measure value, you can't tell waste from investment, so you either cut both or keep both.

Pair every owner with an outcome

When you name a budget owner, name the outcome they're on the hook for as well. Tickets resolved, contracts reviewed, pull requests merged, hours saved. Spend goes on one side of the report, output on the other. The owner's job is to keep the ratio moving the right way.

A Name Next to Every Number

I'll be honest about what I don't know. I'm not sure yet whether AI spend should be charged back to each team's P&L or just shown back to them. Chargeback creates real accountability, and it also makes teams nervous about experimenting at the exact moment you want them to. There are good arguments both ways, and I suspect the right answer depends on how mature your usage is.

What I'm sure of is the starting point. Somebody has to own it. A real person, with a budget, an outcome and the data to explain both.

If you're not sure who owns your AI bill today, start there. Write down the four roles, put names in them, and check whether the budget owner could explain last month's spend by team and workflow. If they can't, the gap is attribution, and that's what our Token & Usage Analytics work sets up: spend mapped to teams, projects and workflows, budgets and alerts by team, and a monthly review that ends with specific actions. Once the owner can see the spend, Token Optimization is the next step. And if you want to talk through how other companies are splitting this up, Book a Discovery Call.

FAQ