Annual budget or rolling forecast for IT spend? Plan the fixed lines, forecast the volatile ones
In brief
When a fixed annual IT budget works, when a rolling forecast is better, and how to run both: plan headcount and contracts, forecast cloud and AI by driver.
Steve Elliott · · 7 min read
On this page
For most IT organizations the answer is both, split by line. Keep an annual operating plan for the spend that is fixed by headcount, contracts and capital approvals, and run a driver-based rolling forecast for cloud and AI consumption, which can move sharply within a quarter.
The mistake, in our view, is picking one method for the whole technology budget. A rolling forecast adds little to a three-year software contract. An annual number set in October tells you little about AI usage, which Bain found grew 4.5 times in a year.
When does a fixed annual technology budget still make sense?
An annual budget still does jobs a forecast can't. It gives the board a spending commitment and gives HR and procurement the authority to hire and sign. Writing for the Association for Financial Professionals (AFP) in 2019, Carl Seidman describes rolling forecasts as a complement to the annual budget, not a replacement: the budget still turns strategy into financial terms and is the plan that actual performance is tracked against.
It also fits the parts of IT spend that barely move once the year starts:
- Staff in seat. Personnel averages 30.8% of IT operating budgets in Avasant's 2025/2026 benchmark, a survey of more than 300 IT organizations, down from 40.9% a decade earlier. In our view, once hires are made the cost is predictable.
- Contracted software and services. Subscription, license and outsourcing fees are set by contract, often for several years.
- Approved capital. Capital budgets average 13.0% of IT budgets in the same study. The money is approved once, then spent on a schedule.
In AFP's 2026 FP&A benchmarking survey of 332 practitioners, only 43% used rolling forecasts; most relied on current-year forecasts. The average budget took 8.7 weeks to build.
When does a rolling forecast work better?
A rolling forecast works better when the main cost drivers change faster than the calendar. A fixed budget is then wrong within months, and the question for leaders becomes where spend is heading, not how it compares with an October guess.
Gartner's forecast for 2026 worldwide IT spending growth was 9.8% in October 2025, 10.8% in February 2026, 13.5% in April and 14.2% in July. Most of the revision came from data center systems, where forecast 2026 growth went from 19% to 62.5%. Gartner ties that mostly to AI infrastructure, much of it hyperscaler capacity rather than a typical enterprise IT budget. Still, in our view, if forecasts of the market moved that much in nine months, an AI or cloud line set last autumn deserves the same skepticism.
The Beyond Budgeting movement goes further. The Beyond Budgeting Institute's principles call for resources made available as needed, "not through detailed annual budget allocations." A BCG article co-written by Bjarte Bogsnes argues the budget tries to serve three incompatible goals at once: stretch targets, unbiased forecasts and resource allocation. Its fix is to separate them, with forecasting as its own process; it notes many companies forecast quarterly over a five-quarter horizon. Deloitte says many organizations are starting to move to a 12- to 18-month rolling forecast.
McKinsey, arguing for reform rather than abolition, links forecasting to reallocation: its 2026 article on fixing the annual budget says top performers shift 10% to 20% of capital year over year, increasingly in-year, and use rolling forecasts to detect when conditions change.
What are the hybrids, and what do "3+9" and "6+6" mean?
Most teams that drop a single annual number land on one of these hybrids:
| Approach | What it is | Horizon and cadence | Best fit in IT | Weak spot |
|---|---|---|---|---|
| Fixed annual budget | One approved number per line for the fiscal year | 12 months, set once | Staff in seat, contracted software, approved capital | Stale by mid-year for consumption lines |
| AOP plus quarterly reforecast ("3+9", "6+6", "9+3") | Plan stays as the target; actuals to date plus a new forecast for the rest of the year | Ends at fiscal year-end; updated each quarter | Whole IT budget where leaders want a fixed target and an honest outlook | Horizon shrinks to three months by Q4 |
| Rolling forecast | Forecast extends a fixed distance ahead each cycle | Typically 12 to 18 months or five quarters; monthly or quarterly | Cloud, AI, hiring plans that cross year-end | Needs reliable actuals and discipline to keep it separate from targets |
| Driver-based planning | Cost is modelled as volume times rate from a few operating drivers | Any horizon; refreshed as drivers change | Usage-priced cloud, tokens, per-seat software | Only as good as the driver data and the rate assumptions |
"3+9" means three months of actuals plus nine months of forecast, produced after the first quarter closes. "6+6" is the half-year version. The plan itself does not change; the reforecast shows where the year will land against it.
The FinOps Foundation's forecasting capability points the same way. At its Run level, forecast models are "a combination of rolling, trend-based, and driver-based" approaches, and one of its success measures is intermediate forecasts that update budgets.
What changes for IT specifically?
Four features of technology spend decide which method fits each line. This is how we would split them:
| IT line | How it behaves | Method we would use |
|---|---|---|
| Staff and contractors | Largely fixed once hired; moves with hire dates and attrition | AOP, with headcount reforecast quarterly |
| Multi-year software and cloud commitments | Fixed by contract; AWS Savings Plans, for example, commit to a set amount of compute usage per hour for one or three years | AOP, with renewal dates tracked in the forecast |
| Capital projects and hardware | Approved once; timing slips and unit prices move | Annual capital budget, with quarterly reforecast of timing and price |
| On-demand cloud and AI tokens | Volume-driven; can change sharply inside a quarter | Monthly driver-based rolling forecast |
Headcount. In our view, the biggest planning risk in staffing is timing, not rate. Ten engineers at an illustrative fully loaded $150,000 who start in July instead of April cost $375,000 less this year. A quarterly reforecast catches that.
Multi-year contracts. Commitments made for one to three years should sit in the plan as fixed. What the forecast needs is the renewal calendar, because that is where the next year's number is decided.
Capex timing. Gartner says rising memory costs are lifting average selling prices for devices in 2026. A hardware refresh approved at last year's prices may cost more or slip into the next year, which changes both cash and depreciation.
Cloud and AI consumption. Bain found the price per token fell by half between December 2024 and December 2025 while tokens consumed grew 4.5 times. It adds that effective cost per task often stays flat, because teams move to better models and run longer tasks. Bain recommends a dedicated AI compute budget instead of funding tokens from existing lines. In our view that budget should be a driver-based forecast, not a fixed figure. Our post on AI initiative costs covers why tokens are still a smaller line than the labor around them.
Illustrative example: one AI line, two forecasts
Take an illustrative $100 million IT budget. Staffing is $38 million, software contracts $22 million, cloud $18 million, capital $8 million, other services $10.4 million and AI model usage $3.6 million.
The AI line was planned as 2 million agent tasks a month at $0.15 a task: $300,000 a month, $3.6 million a year. In the first quarter the agent rolls out to a second business unit. Tasks run at 2.0 million, 2.6 million and 3.5 million a month. Cost per task holds at $0.15.
- Annual budget view. Q1 spend is $1.215 million against a $900,000 plan, a $315,000 variance. On a straight-line view the year looks about $1.3 million over.
- Driver-based 3+9 view. Hold volume at 3.5 million tasks a month for the remaining nine months: $525,000 a month, or $4.725 million. The full year lands at $5.94 million, $2.34 million (65%) over plan, before any further growth.
The driver view also names the lever. Leaders can fund the gap from a reserve, slow the second rollout, or route simple tasks to cheaper models. Bain reports three to five times cost differences between companies that match models to tasks and those that don't. The $38 million staffing line, by contrast, would move by a few hundred thousand dollars if a few hires slipped a quarter. In our view, a monthly rolling forecast there would mostly add work.
How should IT finance choose?
Start from the lines, not the method.
- Sort spend by how it behaves. Mark each line as fixed by headcount, fixed by contract, approved capital, or driven by usage. In the illustrative budget above, staffing, contracts and capital alone are $68 million of $100 million.
- Keep the AOP as the target. The board-approved plan stays fixed for the year, so performance is measured against something stable.
- Reforecast the whole budget quarterly. A 3+9, 6+6 and 9+3 cycle gives leaders an honest landing point without rebuilding the plan.
- Put usage lines on a monthly driver model. Pick the few drivers that explain most of the cost. As a former telecom CFO quoted by McKinsey put it, there are "eight to ten assumptions that influence most of the outcome." For cloud and AI, that is usually volume, unit rate and model mix.
- Extend the horizon where decisions cross year-end. Hiring plans, contract renewals and capital projects need a view into next year by the third quarter, not after the budget kicks off.
Our view is that forecasting works best when it sits in the same operating model as the portfolio and the work, so a change in AI volume shows up as a funding decision, not a variance found in month nine. Where FinOps owns the cloud and AI drivers, agree one number with the TBM office before either reports it.
Sources
- Association for Financial Professionals, AFP Survey Reveals Structured Scenario Planning Separates Top-Performing Corporate Finance Teams, January 20, 2026
- Association for Financial Professionals, Corporate Budgeting Is Becoming Continuous, Connected and More Strategic, October 6, 2026
- Association for Financial Professionals, Carl Seidman, 8 Steps for Creating a Rolling Forecast, September 20, 2019
- Beyond Budgeting Institute, The Beyond Budgeting principles
- BCG, Sebastian Stange, Bjarte Bogsnes and Hardik Sheth, Going Beyond Budgeting, September 13, 2021
- McKinsey & Company, Your budget is killing your strategy—here are four ways to fix it, May 29, 2026
- Deloitte, Yoann Busigny, Implementing the rolling forecast: a 'Beyond Budgeting' approach, December 20, 2022
- Gartner press releases on 2026 worldwide IT spending: October 22, 2025, February 3, 2026, April 22, 2026, July 27, 2026
- Avasant Computer Economics, IT Spending and Staffing Benchmarks 2025/2026: Executive Summary, 2025, and Typical IT Budget Increases Disguise Major Transformation in IT Departments, August 2025
- Bain & Company, How Token Economics Will Change Opex, June 10, 2026
- FinOps Foundation, Forecasting capability
- AWS, What are Savings Plans?
Frequently asked questions
Should IT replace its annual budget with a rolling forecast?
Usually not entirely. The annual budget still gives the board a commitment, sets the benchmark for variances and authorizes hiring and contracts. An AFP article describes rolling forecasts as a complement to budgets. A common approach is to keep the annual operating plan as the target and add a quarterly reforecast, with monthly driver-based forecasts for cloud and AI.
What is a 3+9 forecast?
A 3+9 forecast combines three months of actual results with a nine-month forecast for the rest of the fiscal year, produced after the first quarter closes. It is followed by 6+6 and 9+3 versions. The original annual plan stays fixed as the target, and each reforecast shows where the year is likely to land against it.
How do you forecast cloud and AI spend?
Model it from drivers rather than last year's run rate. For AI, multiply task or request volume by cost per task, and track model mix separately because premium models cost more. Bain found token prices halved in a year while consumption grew 4.5 times, so price and volume need separate assumptions. Refresh the forecast monthly.
How many companies use rolling forecasts?
In AFP's 2026 FP&A Benchmarking Survey of 332 corporate finance practitioners, 43% of organizations used rolling forecasts. Most still relied on current-year forecasts. The same survey found that the average budget took 8.7 weeks to build, unchanged from three years earlier despite investment in planning technology.
Steve Elliott
Published November 22, 2026
More on TBM & IT Financial Management
Cost pools, towers, showback, unit cost.
Keep reading
From an annual cost model to a living one: what changes and what stays hard
Usage-based cloud and AI spend outruns a TBM model rebuilt each year. What a living cost model means, what stays hard, and where AI agents help and don't.
On the TBM Council's numbers, tokens are 4.5–15% of AI cost. What that means for your budget
The TBM Council puts labor at 50–70% of AI project cost. On its numbers, tokens are 4.5–15% of the total. What the figures show and what to budget for.
Marketing Plan vs Launch Plan: Understanding the Key Differences
A marketing plan covers all marketing for a business or product over a period, often a year. A launch plan covers getting one new product or feature to market around a launch date.
ASU 2025-06 and your IT cost model: what changes before you adopt
ASU 2025-06 drops the stage model for internal-use software. What it changes for IT finance: when capitalization starts, the evidence, and the cost model.