What should a CFO be able to see about AI spend?

01The eight numbers

1. Total AI spend this month, across every provider, every card charge for an AI tool, and the AI features inside licensed software.
2. Spend by team, against each team's allocation.
3. Spend by project, against the business case that approved it.
4. Spend against budget for the year to date, with the forecast at the current run rate.
5. Cost per outcome for the five most expensive workflows: per claim processed, per contract reviewed, per ticket resolved.
6. Committed spend: what agents have used this month that has not yet reached an invoice.
7. The untagged remainder: spend no team has claimed, with the person who owns it.
8. Every increase: each cap or budget raised this month, by how much, and who approved it.

The first four tell the CFO where the money went. The fifth tells them whether it was worth it. The last three tell them whether anyone is in control of it.

02Why most finance teams cannot see them yet

AI spend arrives through shared credentials, so the provider invoice shows a total per key and nothing about the team or the project. It also arrives through routes finance does not group together: model providers on the cloud bill, tools on corporate cards, and AI features bundled into software renewals.

KPMG's Q2 2026 AI Pulse Survey found 26 percent of large US enterprises had full, real-time visibility into what their AI systems cost to operate. The same companies planned an average of $202 million in AI spend over the following twelve months. In New Zealand, Datacom's 2026 State of AI Index found one organisation in five has a dedicated AI budget with clear cost tracking.

Ownership is the weak point. In research by the cost-management firm Harness among 700 FinOps and engineering leaders, reported by CIO Dive in July 2026, more than half said their organisation lacks a dedicated owner for AI costs. Roughly one dollar in four spent on AI goes to waste, the same research found.

The result lands on the CFO's desk. The Tokenomics Foundation's State of Tokenomics, published in September 2026 from 472 large companies, found three in four cannot confidently prove AI business outcomes to the CFO.

Budgets are rising into that gap. Forrester's 2027 Budget Planning Guides, published in July 2026, report that more than 80 percent of leaders expect their overall budget to increase in 2027.

03What other firms spend

A useful outside reference is per employee. The Federal Reserve Bank of Atlanta reported in May 2026 that US firms spent $1,358 per employee on AI in 2025 and expected to spend $2,068 in 2026. The figure covers software, hardware, training and support as well as model use, and more than half of firms expected to spend no more than $200 per employee. Treat it as context, and compare your own year on year.

04Where the numbers come from

Numbers one to four need every model call attributed to a team and a project when it runs, which means tags in the path the call takes. Number five needs the workflow recorded on the same line, and a count of outcomes from the system the workflow serves. Number six needs the ledger to be written as the work happens, ahead of the invoice. Numbers seven and eight need a rule that untagged spend has an owner, and caps that cannot be raised without a name attached.

05Questions to ask the CIO this quarter

1. Can we produce the eight numbers for last month, and how long would it take?
2. Which AI spend reaches us outside the cloud bill?
3. Who approved the last three increases, and what did they approve?
4. What is our cost per outcome for the workflow we spend most on, and is it falling?
5. If an agent ran all weekend, at what point would it stop?

06Who owns the number

Finance owns the budget and the report. The CIO owns the path the calls take and the caps in it. Each team lead owns their allocation. The arrangement works when each of them can see the same ledger at the same time, and stalls when finance learns about spend from the invoice and technology learns about the budget from finance.