What a CFO should actually ask about AI spend

"How much are we spending on AI?" is the question every CFO asks first, and it's the least useful one on the list. The number by itself tells you nothing about whether it's healthy. $30,000 a month across 200 people doing real work is a great number. $30,000 a month with no idea which teams, which tasks, or which tools is a number waiting to become a bigger, worse number.

Here are the five questions that actually tell you something.

1. Which team spent it, and did they know their budget while they were spending it?

If spend only shows up as a surprise at month end, nobody was managing it. It just happened to them. A budget with a live number and an alert at 80% turns spend into something a team owns, not something Finance discovers.

2. What happens to a request we didn't define a flow for?

There should be a good answer. In Verybear, undefined tasks land on a safe default ("General assistance"), get logged, and come back as a proposed new flow once a pattern shows up. If your tool's answer is "it just... runs, somewhere," that's the gap where ungoverned spend and ungoverned risk both live.

3. Can we see the model and effort level without seeing anyone's actual work?

Governance and surveillance are not the same thing, and conflating them is how you end up with a tool nobody wants to use. You want aggregate visibility (spend by team, quality signals, flow usage), not a transcript of what Hannah in Marketing asked about her campaign brief.

4. If we tighten something today, when does it take effect?

If the answer involves updating a prompt doc and hoping people read it, it doesn't really take effect. Central config that syncs to every connected tool means a policy change (a blocked request type, a new budget, a model swap) is live everywhere within minutes, not "eventually, for whoever remembers."

5. Who approves a change before it goes live everywhere?

Staged rollout exists for a reason: your AI team should be able to test a flow change against one team before it reaches the other 500 people. "We changed it and found out it was wrong at scale" is not an acceptable answer in any other system your company runs on. It shouldn't be one here either.

Ask these before you ask for the number. The number will make a lot more sense once you do.