An IT budget forecast can have a surprisingly short shelf life.
The annual IT budget may have been approved only a few months ago, but cloud consumption is already running differently than expected, a vendor has changed its pricing, an AI initiative that looked modest during budget season is growing quickly, hiring has shifted, one project moved forward, and another slipped. None of that necessarily means the original forecast was poorly built, just that the assumptions behind it are aging.
That is one of the harder realities of IT financial management. Budget planning processes tend to run on a schedule, while technology spending changes whenever the business, consumption patterns, contracts, or delivery plans change. Spreadsheets can still handle the arithmetic perfectly well; the bigger challenge is keeping assumptions current, spotting meaningful changes soon enough, and understanding how a change in one area affects the rest of the IT budget.
AI has a useful role in that process because it can keep examining historical patterns and new actuals as they arrive, helping ITFM teams identify where the budget forecast is beginning to drift and where a closer look may be warranted. That becomes particularly interesting in areas such as cloud and AI, where consumption can move quickly and some of the cost drivers are still unfamiliar to finance teams.
AI costs themselves are a good example. Token consumption can change dramatically based on user adoption, model choice, workload type, application design, or a single enterprise initiative that suddenly encourages thousands of employees to start using an AI service. Forecasting that with a static annual assumption is difficult, which is exactly why AI-supported budget forecasting may prove useful.
Using AI as an Insight Layer for Cost Transparency: Read Article
IT Budget Forecasting Gets Harder as Technology Spend Becomes More Variable
Some technology costs remain relatively predictable, while others can change quickly enough to make an annual budget assumption feel old within a few months. Cloud has already taught ITFM teams that lesson. Usage can grow faster than expected, and the financial effect may not become apparent until enough consumption has accumulated to show up clearly in the numbers.
AI adds another layer. The direct costs may include model access, token consumption, platform fees, and licenses, but that is only part of the picture. Increased AI usage can also drive more compute, storage, data movement, infrastructure capacity, and supporting services. If the organization rolls out a new AI capability broadly, the cost curve may change before anyone has enough history to know what a “normal” month should look like.
One organization saw token consumption increase by roughly 300% during the week of an enterprise AI training push that encouraged employees to start using a newly available AI tool. The exact percentage matters less than the speed of the change: once adoption accelerated, the prior usage pattern stopped being a reliable guide for the budget forecast.
That is what makes AI spending so difficult to forecast today. There is not yet a long, stable history in many organizations, and usage patterns are still being created. Some employees experiment heavily, some barely use the tools, new applications can generate automated demand at a very different scale, and model pricing itself may change over time. ITFM teams are being asked to budget for a cost category while the organization is still learning what actually drives it.
Vendor pricing, workforce costs, project timing, and application adoption add their own variables. A budget forecast built around five familiar assumptions can suddenly need a sixth or seventh because something that was immaterial last year has become a meaningful cost driver this year. A more adaptive process keeps checking whether the assumptions behind the budget still match what is happening rather than waiting for the next formal forecasting cycle to discover that they no longer do.
Where AI Can Help Keep the Budget Forecast Current
Historical actuals contain a great deal of information about how technology costs behave over time. Cloud, labor, infrastructure, licensing, applications, and services all develop patterns, and AI can help review those patterns alongside current activity to identify areas where the budget forecast may be losing touch with reality.
Perhaps cloud usage has been above forecast for several periods in a row. A vendor category expected to remain flat may have begun trending upward, an application rollout may be generating demand sooner than planned, or token consumption may be increasing much faster than the assumptions used during budget season while the related compute and storage impact starts to appear elsewhere in the technology cost model.
Those findings do not automatically tell the team what the year-end number should be, but they are useful signals that a budget assumption deserves another look.
The same capability becomes valuable for scenario modeling because ITFM teams are constantly asked some version of “What happens if?” What happens if cloud demand grows faster than expected? What if AI workloads increase by 20%? What if token consumption doubles? What if a project starts earlier, a vendor contract changes, or adoption is lower than planned?
The arithmetic behind those questions is often manageable. The time-consuming part is identifying which costs are affected, how those costs relate to one another, and how far the impact carries through the IT budget. AI can help work through those relationships more quickly, compare them with historical behavior, and give the team a starting point that can then be tested against what is known about the business and the technology environment.
Rolling forecasts follow naturally from this. As actuals arrive and conditions change, material differences can be surfaced earlier so the team can decide whether the current budget forecast still makes sense or whether assumptions need to be adjusted.
The Value of a Budget Forecast Depends on When You Find Out
A forecast is most useful while there is still time to do something with it.
Suppose cloud spending is trending toward a significant year-end budget shortfall. If that becomes obvious late in the year, most of the conversation will be spent explaining how the organization got there. Finding the same risk several months earlier leaves more options: capacity plans can be revisited, budgets can be rebalanced, contracts can be reviewed, investments can be reprioritized; if nothing else leadership can at least be prepared for the financial impact.
Consider a cloud forecast that points to a 14% increase over the next 12 months and an $850,000 potential budget shortfall. In another case, a customer-facing application is adopted faster than expected, pushing cloud usage 18% above plan and creating a potential $600,000 variance. The value in both cases comes from seeing the risk early enough to respond, rather than discovering it after the spend is already committed.
That extra decision time can matter more than improving forecast accuracy by a few percentage points. A better question is whether the process helps the team recognize a meaningful change sooner than it would have otherwise, understand the likely financial consequence, and bring that information into the right conversation while there are still choices available.
For many ITFM teams, that would be a meaningful improvement over discovering the problem during a budget variance review after most of the spend has already occurred.
Scenario Modeling Needs More Than a Percentage Change
A question such as “What would a 20% increase in AI workloads do to next year’s IT budget?” sounds simple enough, but the financial answer depends on where that growth actually lands.
Higher AI demand may affect compute, storage, licensing, token consumption, and capacity, although the impact is unlikely to be identical across every category. One area may have enough available capacity to absorb additional demand for a period, while another may be close to a point where new infrastructure or licensing is required. Token costs may increase almost immediately, while other costs move later or only after a usage threshold is crossed.
The useful part of scenario modeling is therefore not applying 20% to everything related to AI. It is using the history and relationships already captured in the ITFM model to understand which costs are likely to move, where the pressure could appear first, and what impact that may have on the budget.
For example, a modeled 20% increase in AI workloads might translate into approximately $1.2 million of additional annual cost, with compute, storage, and licensing among the major drivers and additional capacity likely to be required within nine months. The point of the exercise is not the precision of the $1.2 million estimate by itself; it is the ability to connect a business assumption about AI adoption to likely infrastructure requirements and their impact on the IT budget.
That is much closer to the way budget decisions actually work. Leadership rarely asks only whether spending will rise. They want to know when it will rise, which budgets will absorb it, whether capacity needs to change, and what other investment may need to move as a result.
AI Can Use More Than One Budget Forecasting Method
Forecasting is not one mathematical technique.
A straightforward trend may be well suited to linear regression. Other cost categories may require time-series approaches that account for seasonality, changing growth rates, or recurring patterns. In cases with greater uncertainty, teams may want to explore a range of possible outcomes rather than one point estimate, which is where techniques such as Monte Carlo simulation can become useful.
That does not mean every ITFM team suddenly needs to become a statistics department. In many situations, a relatively simple model will do the job, and adding complexity for its own sake usually creates more work than value. The advantage AI brings is the ability to test different approaches more quickly, compare how well they fit the available history, and recalculate scenarios as new actuals arrive.
A person working manually in a spreadsheet can do this too, but usually not at the same speed or across the same number of variations. Working together, AI can examine patterns, fit alternative forecasting methods, and run repeated scenarios while the practitioner focuses on whether the result makes operational and financial sense.
This may be especially useful for AI costs themselves because the historical record is short and behavior is changing rapidly. Rather than pretending there is one dependable growth rate for token consumption, a team could model several adoption patterns and understand what each would mean for the IT budget. That is a more realistic way to plan for a category where nobody can credibly claim to know exactly what usage will look like twelve months from now.
Keep the Budget Assumptions Where People Can See Them
Every IT budget forecast contains judgment, whether it is built in Excel, an ITFM platform, or an AI-supported planning process. Someone decides which historical periods are representative, whether a recent spike is temporary, which operational measures matter, how growth should be treated, and when the forecast should be rebased.
AI may help with the analysis, but those choices do not disappear simply because the output arrives more quickly. In fact, the finished quality of AI-generated analysis makes it more important to keep the assumptions visible because a clean chart and a well-written explanation can make an inference look more certain than it really is.
Teams should be able to see which history informed the projection, what current signals changed the outlook, and where the analysis depends on an assumption rather than an observed fact. That makes it easier to challenge the forecast constructively.
If token usage increased sharply because every employee was encouraged to experiment with an AI tool during a training week, carrying that growth rate forward for twelve months would probably produce a poor budget forecast. If the increase came from a new production workload that will run continuously, treating it as a one-time spike would be equally misleading. The mathematical model cannot settle that question by itself because the answer depends on what actually happened inside the organization.
ITFM Context Matters in Budget Forecasting Too
A general-purpose AI model can find patterns in a dataset, but useful IT budget forecasting depends on more than pattern recognition. The system needs enough context to understand how technology costs are organized, how applications and services relate to spending, how budgets are structured, which operational measures influence financial outcomes, and how the organization has modeled those relationships.
Historical data becomes far more useful when it retains those connections across cloud, infrastructure, labor, licensing, applications, and services. AI forecasting based on an isolated spreadsheet may detect that one number is rising; an ITFM-informed model has a better chance of showing what that increase affects elsewhere in the budget and why it matters financially.
The data does not need to be perfect before the team begins. It needs to be accurate and complete enough for the budget decision being made, with a clear understanding of where assumptions or gaps remain. That distinction matters particularly with AI spending because the organization may not yet have the mature allocation drivers it has for older technologies.
Token usage may initially be visible only at a platform or account level, then become more granular as tagging, application ownership, model usage, and business consumption mature. The budget forecast can improve along with the data rather than waiting for a perfect model before any analysis begins.
AI Costs Are a Good Place to Experiment—Because They Are Hard
AI spending is an interesting place to start precisely because few organizations understand its future behavior well yet.
Begin with what is known. That might include current token consumption, model or platform charges, recent adoption trends, known production workloads, infrastructure costs associated with AI activity, and any planned deployments that are likely to change demand. Then identify what remains uncertain: employee adoption may continue to increase, a new application might dramatically increase automated token usage, model selection may change the cost per interaction, and infrastructure demand may grow at a different rate from token consumption.
Rather than forcing all of that uncertainty into one budget forecast, build several plausible cases. A baseline could reflect current adoption, another could assume faster employee or application growth, and a higher-demand scenario could show what happens if AI becomes embedded into more production workflows.
AI can help analyze the limited history, look for emerging patterns, and run those scenarios quickly. The ITFM team still has to judge whether the assumptions make sense, but that is exactly why the exercise is useful: forecasting AI costs forces the organization to identify what it actually knows about its AI consumption, what it expects to happen next, and where the biggest unknowns remain.
Better IT Budget Forecasting Creates More Time to Decide
IT budget forecasts are built using the best information and assumptions available at a particular point in time, and some of those assumptions will inevitably change. Cloud already demonstrated how quickly a new consumption model can disrupt traditional budgeting, and AI may accelerate that challenge as token usage, infrastructure demand, and adoption patterns become larger parts of the technology budget.
AI gives ITFM teams another way to notice those changes earlier, test a wider range of possible outcomes, and understand their budget implications before a variance becomes unavoidable. Its ability to analyze patterns and run different forecasting methods quickly can remove a great deal of manual effort, but the value still depends on ITFM context, visible assumptions, and people who understand what is happening operationally.
The opportunity is an IT budgeting process that stays closer to current financial and operational activity and gives leaders more time to decide what to do when the picture changes. For ITFM teams, that is a much more practical goal than trying to produce one perfect forecast of an increasingly unpredictable technology environment.
Confessions from Spreadsheets Anonymous (and the way out): Access eBook
