The FP&A Process: Steps, KPIs and How to Forecast Cloud Costs

Spread the love

Updated November 2026.

The FP&A process is the repeating cycle finance teams use to turn data into decisions: collect and validate data, forecast, budget, compare actuals with plan, and report. FP&A stands for financial planning and analysis. The cycle itself has barely changed in decades. What has changed is the spending it has to plan for: cloud, AI and usage-based SaaS move with customer activity and engineering decisions, and an annual budget built from last year’s run rate misses them by a wide margin.

This guide walks through the five steps, the KPIs worth tracking, what the 2026 surveys say about where FP&A teams struggle, and a worked example of forecasting cloud spend from drivers.

Where FP&A teams stand in 2026

The surveys agree on one theme: data and tooling lag behind ambition.

FindingSource
58% of teams cite data quality and availability as their top bottleneck; 73% rank it in their top three technology challengesVena, 2026 FP&A Impact Report
52% of organizations report revenue forecast variances above 6%Vena, 2026 FP&A Impact Report
70% have executive mandates to adopt AI in finance, yet more than half report only moderate or limited integration between FP&A tools and source systemsVena, 2026 FP&A Impact Report
61% of finance leaders still name spreadsheets as a primary budgeting and forecasting toolVena
Only 34% fully integrate operational drivers such as unit sales, SKUs and headcount into forecasting modelsVena
Machine learning use in forecasting reached 12%, the highest in the survey’s historyFP&A Trends Group, 2026
59% of finance functions use AI in some formGartner, 2025 AI in Finance Survey
45% of finance AI investments lean toward productivity, 20% toward decision qualityGartner, March 2026 survey of 204 finance leaders

The last row matters most. Gartner warns that this imbalance can create a gap where finance reports AI progress but boards see limited strategic impact. Faster reporting is useful; better decisions are the point.

The five steps

1. Collect and validate data

Gather actuals, operational drivers and assumptions, and check that definitions match across systems. Many forecast errors start with two teams using the same word differently: “active customer” in billing versus product analytics, or “cloud cost” with and without credits and commitments.

For cloud, validate against the billing export, not a dashboard, and decide once whether you report amortized or cash cost of commitments. FOCUS-format exports from AWS, Azure and Google Cloud make this easier; see multi-cloud cost management.

2. Forecast

Build driver-based models with base, upside and downside cases, and refresh them on a rolling basis. Rolling forecasts remain rare in practice: even among self-reported mature teams, Vena found only 11% use agile methods such as rolling forecasts. That is the single biggest process gap for usage-based costs.

3. Budget and plan

Turn strategy into spending authority with an owner for every line. For cloud and AI, the owner should be an engineering or product leader who controls the drivers, with finance owning the method. Cloud cost allocation gives each line an owner.

4. Analyze variance

Compare actuals with budget and forecast, and explain each gap in terms an owner can act on: “volume up 12%, unit cost flat” is actionable; “cloud over budget” is not. Split variance into volume (more customers or usage), rate (price or efficiency changes) and mix (new services).

5. Report and support decisions

Dashboards, management reports and ad hoc models that leaders actually use. The best FP&A reports pair every cost with the unit it serves, so a rising bill can be read as healthy growth or as margin erosion.

KPIs worth tracking

KPIWhat it tells you
Forecast accuracy by lineWhether the planning process works
Revenue growth and gross marginOverall health
Operating cash flowLiquidity
Opex as a share of revenueEfficiency at scale
Unit costs, such as cost per customerWhether growth is profitable
Cloud and AI cost as a share of revenueWhether usage-based costs scale with the business
Commitment coverage and utilizationWhether fixed cloud obligations are used

For forecast accuracy on cloud lines, the FinOps Foundation’s maturity model offers a sample Crawl-stage goal of forecasting within about 20% of actual spend, tightening as the practice matures. See the FinOps maturity model.

Why cloud and AI costs break the model

A fixed cost line is predictable. Usage-based costs are not, for three reasons:

  1. They move with activity. More customers, more requests, more data, more cost, every hour.
  2. Engineers change them daily. A deployment on Tuesday can change next month’s bill.
  3. Prices and pricing models change. In 2026 alone, GitHub Copilot moved to usage-based billing and Cursor moved to API-rate billing for third-party models. See Cursor pricing.

Three adjustments handle most of this:

  • Forecast from drivers, not last month. Tie spend to customers, requests or jobs, and update the cost forecast whenever the driver forecast changes.
  • Separate committed from variable spend. Savings Plans and Reserved Instances are fixed obligations; model them as such. See Reserved Instances versus Savings Plans.
  • Review monthly with engineering in the room. A month-end explanation from finance alone arrives too late to change anything.

A worked example: forecasting cloud from drivers

Take a hypothetical SaaS company planning next year’s cloud budget. Today it has 2,400 paying customers and a cloud bill of $120,000 a month, or $50 per customer.

Run-rate method: $120,000 × 12 = $1.44 million. Simple, and wrong as soon as anything changes.

Driver-based method:

DriverPlan
CustomersGrow from 2,400 to 3,600 by year end (average about 3,000)
Cost per customer, excluding AIFalls from $50 to $46 after rightsizing and commitments
New AI featureAdds about $5 per customer per month from Q2
Committed spend$40,000 a month in Savings Plans, fixed

Average monthly cost: about 3,000 customers × $46 = $138,000, plus AI at $5 × 3,000 for nine months, which averages $11,250 a month over the year. The budget comes to roughly $1.79 million, about 24% above the run-rate number.

That gap is the reason to build the model. The run-rate budget would have been blown by the first quarter of growth, and the variance report would have blamed engineering for spending that the business had planned.

Three scenarios complete it: a downside with slower growth (commitments now look large relative to usage), an upside with faster growth (cost scales but cost per customer should fall), and an AI-heavy case where adoption doubles the AI line.

Tools that help

AWS Cost Explorer now forecasts up to 18 months ahead using up to 36 months of history, and adds AI explanations of what drives the forecast. Treat it as a baseline trend that knows your history but not your roadmap. See the AWS Cost Explorer guide.

Planning platforms connect drivers from CRM and product data to the financial model. FinOps platforms supply allocated cloud costs and unit costs. The common thread in the surveys above is integration: the forecasts that work pull drivers and costs from source systems rather than re-keying them into spreadsheets each month.

Where AI fits

Gartner’s 2026 research suggests most finance AI spending targets productivity: faster close, automated reconciliations, quicker report drafting. The bigger prize is decision quality, such as better driver forecasts and scenario analysis. FP&A Trends Group data shows AI users reporting more time on insight and higher forecast quality, but the majority of teams have not adopted machine learning in forecasting yet.

Start where data is already clean. Cloud billing data, with its hourly granularity and consistent schema, is often a better first forecasting use case than revenue.

FP&A and FinOps

FinOps is, in large part, FP&A applied to technology spend. Finance brings forecasting method and budget discipline; engineering brings the drivers and the levers. Both look at the same allocated, unit-based data. What is FinOps covers the operating model, and FinOps for AI covers the fastest-moving line in most 2026 budgets.

Questions people ask

What are the steps of the FP&A process?

Collect and validate data, forecast, budget and plan, analyze variance, and report to support decisions.

How often should forecasts be updated?

Monthly at minimum, and more often for volatile usage-based lines such as cloud and AI.

What is a good forecast accuracy for cloud costs?

It depends on maturity. The FinOps Foundation’s sample Crawl-stage goal is within about 20% of actual spend.

How should FP&A handle Savings Plans and Reserved Instances?

As fixed commitments, separate from variable usage, with utilization tracked monthly.

Sources

Nhon Dang

Nhon Dang is a cloud infrastructure and operations professional with over 10 years of hands-on experience in cloud services, infrastructure, and business operations. His expertise spans the design, deployment, and operation of cloud platforms and managed services, including virtual machines (VMs), Kubernetes (K8s), object storage (S3), managed databases, Apache Kafka, and cloud GPU infrastructure. Throughout his career, Nhon has worked closely with cloud infrastructure and service operations, gaining practical experience in building reliable, scalable, and cost-efficient cloud environments. His work combines technical expertise with business and operational insight, giving him a practical perspective on how cloud technologies perform in real-world production environments. Nhon writes about cloud infrastructure, Kubernetes, DevOps, distributed systems, cloud computing, infrastructure operations, and cloud service management, sharing insights based on hands-on experience rather than purely theoretical knowledge. His goal is to provide practical, technically accurate, and experience-driven guidance that helps engineers, technical teams, and businesses make better decisions when adopting and operating cloud technologies.

Leave a Reply