Your forecast will be wrong. That is not necessarily a failure of FP&A. The real failure is allowing the business to keep operating against assumptions that you already know are no longer true. A forecast represents your best view of the business based on the information available today. As customer behavior, hiring plans, pricing, productivity, and technology costs change, the forecast should change with them. AI makes that discipline even more important because it introduces new variables into both the operating model and the cost structure. Revenue assumptions can shift as products change faster, hiring plans can move as productivity changes, and technology costs can vary with usage. Token-based AI costs add another variable because spending can depend on interactions, models, and workflow complexity.
The real FP&A test is whether your finance team can explain what changed, quantify the impact, and help leadership decide what to do next.
That requires a different relationship with the budget. Your budget still matters as a benchmark. But it cannot be your only financial view.
Key takeaways
- A forecast is a set of assumptions, not a promise that the business will follow one path.
- AI increases the need to understand cost drivers, not simply add an AI line to the budget.
- Token budgeting connects AI usage to financial impact.
- Variance analysis matters because the reason for a change often matters more than the size of the change.
- Rolling forecasts keep your forward view current as new information replaces old assumptions.
- Strong FP&A turns changes in performance into decisions about resources, risk, and priorities.
Why does AI make traditional budgeting harder?
AI changes the cost behavior behind some technology spending. Traditional software often gives you a predictable cost based on licenses, users, or contracted fees. AI can introduce consumption-based costs that move with usage.
For example, an AI-enabled product may consume more tokens as customers use a feature more often. An internal agent may consume more tokens as you give it longer context, more complex tasks, or additional steps. Your budget therefore needs to connect spending to the activity that creates it.
You do not need to predict every AI use case twelve months in advance. You need to identify the drivers that matter, set reasonable assumptions, and create a process for updating them as actual usage becomes visible.
According to the IBM Institute for Business Value, a survey of 2,000 executives found that business leaders expect AI investment to increase by about 150% as a percentage of revenue between 2025 and 2030. The planning implication is clear: AI spending is becoming a larger financial consideration, not just an isolated technology experiment.
Your budget should set direction, not freeze decisions
An annual budget answers an important question: What are you planning to do with your resources? A forecast answers a different question: Given what you know now, what do you expect to happen?
The budget is the baseline. The forecast is the latest view. The variance is the information. Those three should not be confused. Management should continue to understand performance against the original plan while also making decisions using the most current view of the business.
Keeping those questions separate makes your management process more useful. If sales are below plan, you do not need to pretend the original budget was wrong. You need to understand whether the variance is temporary, structural, or caused by an assumption that no longer holds.
Consider a company that budgeted $2 million for a new product launch. Six months later, the launch is delayed, customer adoption is slower than expected, and the team has started using AI to automate parts of the product and support process.
The useful question is not, “Why are we missing the budget?” It is, “What changed, what does that change do to our economics, and what should we change now?”
That could mean delaying a hire, moving spend to another initiative, changing the expected margin, or increasing investment because the economics have improved. The right answer depends on the facts. FP&A creates the structure to see those choices clearly.
What should you do when actuals diverge from the forecast?
Start with the driver, not the dollar variance. A $500,000 revenue shortfall caused by lower customer volume requires a different response from a $500,000 shortfall caused by pricing or timing.
A material variance should be decomposed into the operating drivers behind it. For revenue, that might mean volume, price, mix, conversion, retention, or timing. For AI costs, it might mean adoption, usage intensity, model mix, pricing, or workflow efficiency. This decomposition helps FP&A distinguish between a change in business activity, a change in economics, and a change in timing before determining what action management should take.
Your variance review should answer four questions:
- What changed?
- Why did it change?
- Is the change temporary or likely to continue?
- What decision should management make because of it?
This is where FP&A becomes a management function rather than a reporting exercise. The value is not in producing another variance report. The value is in connecting the variance to an action.
A practical review might show that cloud and AI costs are 20% above forecast. The variance decomposition tells you what drove the increase; the next step is determining what that driver means for the business.
Each explanation leads to a different decision. Higher usage may support more investment. Inefficient workflows may require redesign. Uncontrolled experimentation may require spending limits. A pricing change may require a new forecast assumption.
How should you build a token budget for AI?
Token budgeting is useful, but tokens should not become the financial metric management optimizes in isolation. Finance should connect AI consumption to the business activity generating the cost. That means translating token usage into operating metrics such as cost per customer interaction, cost per transaction, cost per workflow, cost per developer, or cost per unit of revenue generated. The question is not simply, “How many tokens are we consuming?” It is, “What business outcome are we purchasing with that consumption?”
A useful token budget starts with the business activity that creates the cost. Instead of budgeting one large AI expense line, connect expected consumption to specific uses.
For example, you might budget separately for:
- Employee productivity tools.
- Customer-facing AI features.
- Software development and coding workflows.
- AI agents that perform multi-step tasks.
- Research, testing, and experimentation.
Then connect each category to a measurable driver. That could be users, transactions, documents processed, customer interactions, or another unit that reflects actual use.
A practical AI forecast can take this one step further by breaking cost into four core drivers: Volume × Usage Intensity × Model Cost × Efficiency.
Volume captures how many users, customers, transactions, or workflows are using AI. Usage intensity captures how much AI each activity consumes. Model cost reflects the underlying model mix and pricing. Efficiency captures changes in prompts, context, caching, routing, or workflow design. Breaking AI costs into these drivers makes it easier for FP&A to understand not only whether spending is above or below forecast, but why it changed.
This framework gives FP&A a clearer way to explain AI cost variances and determine whether higher spending reflects greater adoption, changing economics, or inefficiency. It also creates a better basis for deciding whether the additional cost is justified by the business value being created.
The goal is not to minimize token consumption. The goal is to understand what you are buying with that consumption.
What does a rolling forecast change?
A rolling forecast keeps your forward-looking view current instead of allowing it to become stale as the year progresses. IBM defines a rolling forecast as a model that continuously predicts future performance by adding a new period as each period is completed.
For a finance leader, the value of a rolling forecast is not that finance updates every assumption every month. It is that material assumptions are refreshed as new information becomes available so leadership is not making decisions based on an outdated view of the business. The appropriate cadence may be monthly or quarterly, depending on the business and the volatility of its major drivers. But the goal is not constant rebuilding. It is maintaining a current economic view of the business.
Focus updates on the drivers that have actually changed. Customer growth may need to be updated while pricing remains unchanged. Hiring dates may move while compensation assumptions stay intact. AI adoption may accelerate while infrastructure pricing remains stable.
Kranz’s guidance on flexible budgeting similarly emphasizes rolling forecasts, scenario planning, cash flow monitoring, and regular budget reviews as ways to keep financial plans responsive as conditions change.
How can FP&A turn uncertainty into better decisions?
The strongest FP&A teams do not try to eliminate uncertainty. They make uncertainty easier to manage.
That means building scenarios around the variables that can materially change your business. For AI spending, you might model lower, expected, and higher usage. For revenue, you might model different customer growth or conversion rates. For headcount, you might model different hiring dates.
The point is not to create dozens of scenarios. Too many scenarios create noise and slow decisions. Focus on the few assumptions that could materially change cash, profitability, runway, or your ability to hit a strategic milestone.
You should also define what would cause you to move from one scenario to another. The most useful scenarios have defined triggers that connect a change in performance to a potential management response. If AI usage exceeds a defined threshold, management might review model selection or workflow design. If revenue falls below a defined level, the company might pause discretionary hiring. Defining these triggers in advance turns scenario planning from a modeling exercise into a decision framework. Management knows what to watch, when a change requires attention, and which actions should be considered if the underlying assumptions move.
AI can improve forecasting without replacing FP&A judgment
AI can help finance teams analyze data, identify patterns, generate forecast narratives, and reduce manual work. Used thoughtfully, it can streamline planning and forecasting while giving finance teams more time to focus on higher-value analysis.
The important distinction is that automation does not remove judgment. AI can identify that marketing spend is above forecast, but determining whether that variance reflects a successful growth investment, inefficient spending, or a timing issue still requires business context.
As AI automates more of the mechanics of forecasting, the value of FP&A shifts further toward interpretation, challenge, and decision support. Finance still needs to determine which assumptions matter, understand why performance changed, challenge the operating narrative, and translate financial outcomes into management decisions.
AI can accelerate the analysis. FP&A provides the judgment.
A better budget process starts with better questions
If your forecast changes every quarter, that does not mean your planning process is broken. It may mean your finance team is doing what it should: incorporating new information.
The warning sign is different. You should be concerned when your forecast changes and nobody can explain why, when large variances repeat, or when AI spending grows without a clear connection to outcomes.
A strong FP&A process creates a consistent discipline around identifying what changed, understanding the financial impact, and determining whether management needs to respond.
That is more valuable than producing a forecast that happens to be right for the wrong reasons.
Frequently asked questions
CFOs should budget AI costs based on measurable usage drivers rather than treating AI as a fixed software expense. Useful drivers include users, transactions, AI interactions, documents processed, workflows, and model usage, with separate assumptions for experimentation and production workloads.
Token budgeting is the process of estimating, allocating, monitoring, and managing the AI tokens consumed by specific users, features, workflows, or agents. It helps finance connect variable AI usage to financial costs and evaluate whether the resulting business value justifies the spend.
Yes. AI spending should be included in the annual budget, with material changes in usage, pricing, model mix, or adoption reflected in updated forecasts throughout the year.
Rolling forecasts allow finance to replace outdated AI cost assumptions with actual usage trends and maintain a current view of expected spending.
FP&A can improve AI investment decisions by connecting AI spending to measurable business drivers and outcomes. Instead of asking only how much an AI tool costs, finance can evaluate cost per user, transaction, workflow, or other relevant unit alongside revenue, productivity, margin, or capacity outcomes.
What should your FP&A process look like in the age of AI?
Your FP&A process should treat the budget as a baseline and the forecast as a living view of the business. The goal is not to make every forecast accurate months in advance. The goal is to identify meaningful changes early, understand what caused them, and give leadership a clear set of choices.
That may mean strengthening your financial planning and analysis process or adding more structured CFO advisory and strategic finance support as your business becomes more complex. Kranz can help you build the models, reporting cadence, scenario analysis, and decision support needed to keep your financial plan connected to what is actually happening in the business.
Your forecast will change. The question is whether your organization can understand why it changed and act quickly enough for that information to matter.
Turn your forecast into a management tool
A useful forecast should do more than tell you where you may land. It should help leadership understand what changed, what it means for the business, and what decisions should be considered while there is still time to act.
The goal is not perfect forecast accuracy. It is decision readiness.
If your budget, forecast, and operating decisions are no longer moving together, Kranz can help you build a more responsive FP&A process.
Insights By
