Planning season action plan for finance leaders: five posts on AI spend

It’s budget planning season, and most finance teams find AI spending the hardest part of the budget to estimate. It shows up across the P&L, from COGS to R&D to sales and G&A. The provider invoice only tells you the total and not much more.

Every other line in the plan has a known driver and a named owner. Our VP of Finance, Dan Carducci, described the standard in his post last week (“AI finally plans like every other line in my budget“). For each core element of his plan, “I can name the driver and the owner and tell you how far the number moves when the assumption changes.” Most AI budgets can’t meet that standard yet.

Over the last few months, members of our finance team and others at CloudZero have written about this problem. We picked five of those posts. If you read them in order, they take you from the invoice to a plan you can defend.

1. Why the AI bill can’t answer your planning questions

From a $60K invoice to a $200B earnings call, few can explain the AI bill

AI providers bill tokens by model and by day. The invoice doesn’t show which tokens built a shipped feature and which went to an old workflow nobody moved to a cheaper model. The post reconciles a $61,656 sample invoice by CJ Gustafson of Mostly Metrics. Once usage was matched to the bill, the waste came from outdated model choices, unmanaged caching, paying for faster processing than the work needed, and a few unreviewed jobs that drove most of the spend.

It also draws on CloudZero’s AI Economics Pulse. Across 430 customer organizations, the median company’s AI spend reached 2.66% of its cloud bill in August, up from 0.67% a year earlier. At the 75th percentile, AI passed 10% for the first time.

What to take into your plan: Give every category of token spend an owner, and pair each cost with a value metric before you cut anything. Finance teams that see AI spend within a day are more than twice as likely to hold an “invest aggressively” posture.

2. The decisions to make before the number locks

Your FY27 plan deserves a real AI number, not a hedge

A plan needs AI spend by customer, product, team, and P&L category. The monthly bill only shows totals by vendor and service. So the AI number in most plans becomes a cushion that gets trued up later, often more than once.

In CloudZero’s June 2026 survey of 260 senior finance leaders, 87% said they need to tie AI spend to business results within the next year, and 22% can do it today. Among teams that can’t measure AI outcomes, 75% have held back on AI investment and 35% have killed an initiative.

What to take into your plan: Answer three questions before the plan locks. Is AI a risk or an opportunity in your plan? How does a dollar of AI compare with a dollar of headcount or software? Are you capitalizing any AI-assisted development work, or is all of it going to OpEx?

3. Build the view that connects AI to the business

The finance dashboard I actually use, built from CloudZero and Campfire in an afternoon

Emily Allen, our Director of Finance and Accounting, doesn’t write code. She connected Claude to CloudZero and Campfire through their MCP servers and built the dashboard she now uses every week. It covers the P&L, customer margins ranked by dollars, and AI spend as a share of revenue, gross margin, and salaries by department. It also has the model mix, a capitalization view under ASC 350-40, and a sensitivity table for levers like moving a quarter of premium-model usage to a cheaper model.

She reconciles every result to her own reporting, which she treats as the control. The post includes her prompts in order. She estimates the dashboard saves her five hours a week.

What to take into your plan: Measure AI against revenue, margin, and payroll by department to see where to look before you set targets. Use the sensitivity levers in your scenarios.

4. Let budget owners make their own case

Our Customer Success AI bill tripled. Here’s why we’re spending more.

Scott Castle, our chief product officer, starts with the problem of averages. Split a $40,000 monthly Anthropic bill evenly across two customers who each pay $25,000, and each returns $5,000 in margin. Allocate it by the workload each customer drove, and one costs $32,000 and loses $7,000 a month, while the other costs $8,000 and returns $17,000.

Matt Katz, our SVP of Global Customer Success, applied the same approach to his team. Its AI spend per employee tripled from March to July. He tied that spend to higher platform engagement and a better renewal rate, and renewal revenue came to four to five times the AI spend. The post includes the webinar recording and full transcript.

What to take into your plan: Allocate AI spend by the activity each customer or team drove, since averages hide the ones losing money. Ask each budget owner to bring their 2027 AI request with the outcome it’s tied to.

5. What the finished plan looks like

AI finally plans like every other line in my budget

Dan Carducci, our VP of Finance, explains how CloudZero’s 2027 AI line came together.

Six months ago, his AI forecast was the run rate plus a cushion, and every AI budget question went to him because only he could see the total. That changed once his team could break the spend down. Emily traced a doubling of AI costs to one account development rep’s heavy premium-model use. Moving part of it to a cheaper model was worth $13,000 to $30,000 a month, and the analysis took her 10 minutes. Last quarter, the team also capitalized the AI spend that went into new feature development.

Each team now plans around its own cost per unit of work, such as one customer review for customer success or one close for sales. Budget owners see their AI numbers the same way Dan does, so 2027 requests arrive as cases he can evaluate like any other.

What to take into your plan: Build the AI line with the people who own the spend, around cost per unit of work, and model what happens if usage doubles. In our survey, companies were twice as likely to overspend their AI budget as to hit it.

Before you lock your AI number

Taken together, these five posts point to the same questions to answer for the AI line in your 2027 plan:

If you can answer those, the AI line can be planned, reported, and defended the same way as the rest of your budget.

To see how your organization compares, try CloudZero’s AI Benchmark Tool. It asks five questions about how your AI initiatives are driven, tracked, and funded, and shows where you stand against the 260 finance leaders in our survey.

More reading for planning season

Author Spotlight

David Aponovich

David is CloudZero’s senior product marketing manager. He brings a diverse background in technology as a product marketer, industry analyst, and content management advisor, among others. He’s an amateur outdoor photographer and has more photos on his iPhone than you’d believe.