AI spend is one of those categories that landed on the budget before most finance teams had a process for it. Tools got approved, subscriptions piled up, and now the question I hear most often is: why does this number keep moving? Forecasting AI costs is genuinely hard right now, and most companies have not figured it out yet. That is not a failure of planning. It is what happens when pricing models are new, vendor behavior is unpredictable, and adoption inside the company is still accelerating.

This edition is about how finance teams can move from reactive to active on AI spend. Not just forecasting it better, but actually owning the category: setting controls, managing vendor relationships, and running a review process that catches problems before they hit the quarterly close.

The free section covers why AI costs are different to manage and three levers your team can pull now. Paid subscribers get the full playbook with all five templates in a single Excel file, including an AI subscription tracker, credit allocation framework, model routing guide, vendor negotiation checklist, and quarterly review.

CFO Skills for Claude pack now has three editions: Financial Reporting, Cash Flow Management, and Audit and Controls.

The paid section this week has a new one - FP&A and Budgeting.

Today is day one of the first Claude in Action cohort. We have an incredible group in the room, and I am excited about what the next few weeks look like.

The July cohort waitlist is now open. If you want in at the early pricing, sign up here.

If your company needs something more structured, book a call with me. I run a three-session program as a starting point, but in most cases, we build something specific to your team: the tools you are actually using, the workflows that matter, and the gaps that are costing you time right now.

A client of mine runs a software development company. Smart team, disciplined on costs. Earlier this year, they came to me with a question: our AI bill keeps climbing, but our headcount has not changed. Why?

They had been treating AI spend like any other software line. Tools approved, budget set, finance checks in at month-end. That process works for SaaS. It does not work for AI, because the pricing model is different in a way that matters.

This is not a small-company problem. A few recent data points that landed in my feed this week:

The pattern is the same at every scale. The spend arrives faster than the process does.

Most teams start with personal or team subscriptions: a fixed monthly price, a usage limit, and you are either throttled or locked out when you hit it. Predictable costs. Real constraints. When someone reaches their daily cap mid-afternoon, they either stop or find a workaround, which is its own problem.

Enterprise plans work differently. No hard caps, no throttling. You pay for actual token consumption: every input, every output, billed at a rate per million tokens. The model works well for heavy users who need reliability, but the cost behavior is completely different. Two employees doing similar work can generate bills that are several times apart depending on how their tasks hit the model. A developer running long code reviews through a premium model all day looks nothing like a marketer drafting emails on the same platform.

According to Ramp data, the top quartile of businesses by AI spend pays about four times what the median business pays with the same vendor. For some vendors, that gap exceeds 15x. That is not a sign of misuse. It is what usage-based pricing looks like at scale.

And within a single vendor, the model tier is its own variable. Anthropic's most capable Claude model is currently five times more expensive per million tokens than its more basic model. Most employees default to the premium tier for every task because nobody told them not to. That guidance needs to be explicit, not assumed.

Most of what is on the AI bill today was not purchased through a formal process. Ramp data shows AI-related reimbursements tripled year over year. Nearly two-thirds of businesses using AI are already running more than one tool. Someone signed up, billed the corporate card, or got reimbursed. By the time finance sees the full picture, there are overlapping subscriptions across departments and tools nobody owns.

This is the environment you are managing in.

Here is where I want to be direct: forecasting AI costs right now is genuinely hard, and a clean model is not realistic yet for most companies.

Token pricing changes. Vendors reprice models, often without much notice. Usage scales non-linearly as adoption grows inside the company. The forecast that worked for a 20-person pilot looks nothing like what you need when the full organization is on the platform. A budget set in Q1 can miss badly by Q3 without anyone doing anything wrong.

The practical approach is a range with a meaningful buffer, quarterly recalibration, and an early-warning process for when actuals start diverging from the plan. Track actual spend by team, not just total spend. If one department is running hot, you want to know in week three, not at month-end.

Vendor negotiations are also part of the forecasting answer. AI vendors are in a competitive market right now, and the terms are more flexible than most finance teams realize. Short-term contracts protect you when pricing shifts. Pricing escalation caps of 3 to 5 percent give you a ceiling. Credit pooling across the organization and rollover provisions prevent waste when a team underuses a plan. Grace periods in the first few months while you calibrate actual usage are also standard asks in enterprise contracts. None of this is guaranteed, but it is negotiable, and finance should be in that conversation before the contract is signed.

We have covered what makes AI costs different and where the easy controls live. The subscriber section is the operational part: a step-by-step process your finance team can actually run, plus templates to do it without starting from scratch.

This is a genuinely new challenge for finance teams, and I want to be clear about that. There is no clean playbook yet. Most companies are still figuring out how to forecast a cost that changes with every vendor update and scales in ways that are hard to predict. The templates in this edition are a starting point, not a finished answer.

What I do know is that finance is the right function to own this. You already control the budget approvals, the vendor relationships, and the reporting. The missing piece is usually just the process and the habit of treating AI spend as an active category rather than a passive one.

As always, let me know what you are seeing on your end. Are your AI costs behaving? Or are you also getting surprised?

Until next Tuesday.

Anna

This newsletter is for you, and we want to make it as valuable as possible. Please reply to this email with your questions, comments, or topics you'd like to see covered in future issues. Your input shapes our content!

Want to dive deeper into balanced AI adoption for your finance team? Or do you want to hire an AI-powered CFO? Book a consultation!

Did you find this newsletter helpful? Forward it to a colleague who might benefit!

Until next Tuesday, keep balancing!

Anna Tiomina
AI-Powered CFO