Build an AI Budget Around the Work You Intend to Run

Six · 6 min read · August 18, 2026

Build an AI Budget Around the Work You Intend to Run

Say you run a small consulting firm, and at the start of the month you agreed on an amount to spend on AI. Nobody argued about the number. It sounded reasonable, and everyone had ideas for how to use it.

At the end of the month you look at the total. It's close to what you planned, maybe a little over. Then your finance lead asks which work earned that money, and nobody can answer. Some of it went to the brief your operations lead prepares for the Monday meeting. Some went to a client proposal that took four rounds of revision. Some went to experiments nobody remembers starting. You know what you spent. You don't know what you got for it, or which of it should happen again next month.

The number came first, and the work filled it. Start in the other direction. Name the work you intend to run, then decide how much you're willing to fund while you find out whether that work helps.

Why a monthly number tells you so little

Part of the problem is that AI spending is hard to forecast, and the people whose job is managing technology costs say so. The FinOps Foundation, which writes practice guidance for those teams, describes AI cost and usage as new to many organizations. Its guidance says forecasting "may be much more challenging for new technology areas," and that "funding for projects may need to be revisited more frequently until forecast accuracy is improved."

So the figure your firm set at the start of the month was a guess. A guess is fine as long as everyone treats it as one. It goes wrong when the guess turns into the plan, and the only question at month end is whether you stayed under it. Coming in under budget says nothing about whether the Monday brief helped, or whether the fourth round on the proposal was worth paying for.

Budget the assignments, not the promise

"Use AI for operations" isn't a budget. Neither is "give every team access." Neither one tells you what work to expect, who will check it or when you'll decide whether it continues.

Start with the assignments themselves. For the firm, that might be three pieces of work: the brief your operations lead takes into the Monday meeting, the revisions on a proposal your account lead is preparing for a client, and a single research packet the founder wants before meeting a possible partner. Each one gets an owner and a reason to exist.

Then ask the harder question: what result would make the next run worth funding?

For the Monday brief, it's a version the operations lead can read before the meeting and find what the team actually needs to discuss. The proposal has a plainer test: would the account lead check this draft line by line and send it? The founder wants the research packet to show what's still unknown about the partner. None of these depends on a saving you hope to make later, and each owner can check theirs when the work comes back.

A sample month

Here's the firm's month on one page. Your rhythm, owners and questions will be different, so swap in your own.

Intended work Planned rhythm Owner Review question
Operating brief Weekly Operations lead "Did this make Monday's meeting easier to prepare?"
Proposal revision As needed, up to four runs Account lead "Can I see what changed in the scope, and would I send this version?"
Partner research packet One run Founder "Did this narrow my decision, or show me a question we still need to answer?"

There are no amounts in the table yet. The amount still matters, but it's easier to set once you know what each line is for and what its owner will check. At that point you can ask a narrower question about each one: how much are we willing to spend to find out whether this helps?

At the end of the month, sit down with each owner and go line by line. There are four useful answers: keep it as it is, change the assignment, make it smaller, or stop.

Say the Monday brief arrived every week, but your operations lead rewrote the client-deadline section each time because it was pulling from an old project list. The brief is worth keeping. The assignment needs to change, so point it at the current list and see whether next month's versions need less rework. The proposal used all four runs, and the last two only reworded sentences the account lead could have fixed in ten minutes. Next month it gets two. The research packet answered the founder's question and the meeting has happened, so it stops.

Give every line a place to stop

Each line needs a stopping point that comes from the work itself. The Monday brief is finished when the operations lead stops using it to prepare. The research packet is done once the founder has enough to choose the next question. A proposal stops when the account lead accepts a draft, or decides the open scope needs a conversation with the client more than another revision.

When a piece of work has no review point, a spending limit becomes the only thing that can tell it to stop. You should still set limits, because they catch the run that went further than anyone expected. But a limit only knows what the work cost. It had no way to see what the account lead saw in the proposal: the last two runs only reworded sentences.

Leave room for the work you didn't plan

Some of the firm's AI use won't be on the list. A client calls on a Wednesday asking for a quick comparison of two vendors, and your account lead wants help turning it around by Friday.

You don't need to forbid that work. Give it the same treatment the day it starts: an owner, the question they'll ask when it comes back and the point where it stops. If the vendor comparison turns out to be something clients want every month, it earns a line in next month's table. If it was a one-off, it stops when the account lead sends it. Either way, at month end you can say what that money bought, which is the thing the firm couldn't say before.

What the NIST framework asks for

The NIST AI Risk Management Framework is voluntary guidance from the US National Institute of Standards and Technology, written about AI systems and risk rather than budgets. Even so, it asks for the two things the firm's table writes down: that an AI system's "intended purposes" are "understood and documented," and that the roles and responsibilities around it are documented too. It also describes a point where "a determination is made as to whether the AI system achieves its intended purposes and stated objectives and whether its development or deployment should proceed." For the firm, that's the month-end conversation: did the Monday brief do what you wanted, and should it run again?

Where Supanova fits

Say the firm hands those three assignments to Supanova's autonomous workforce. You choose how often your workforce checks in: every step, every task, or once per project. For the Monday brief, that check-in is where the operations lead asks whether the brief made the meeting easier to prepare. Your finance lead gets the money side at month end, in a usage history that leads with what each charge was for and when it happened.

If the proposal revisions are still running when they hit your workspace's spending limit, that work pauses and you decide whether it gets more room. Running out of money is a different stop. If it happens midway through a Monday brief, the brief waits, and the Chief of Staff in your workforce tells you exactly what it needs and asks once.

None of this decides whether the Monday brief deserves next month's money. That decision stays with you and the person who reads it each week.

Before next month starts, pick three pieces of work you actually mean to run. Write down who owns each one, the question they'll ask when it comes back and the point where it should stop. Then set the amount. When you're ready to hand that work to a workforce, start at supanova.team.

Sources

autonomous-workforceAI spendAI budgetAI Operations