To forecast Google Cloud spend well you need three things: a clean baseline of what you already spend, a model of what drives that spend up or down, and a feedback loop that compares the forecast to the actuals every month. Most teams have the first, guess at the second, and skip the third, which is why their numbers drift. The method below fixes all three and produces a forecast that holds up in a board meeting.
This how-to sits in our Google Cloud cluster. The wider context lives in our complete guide to Google Cloud cost optimization, the pillar this piece links up to, and it pairs naturally with our walkthrough on setting budgets and alerts in Google Cloud, because a forecast without a budget to check it against is just a hope.
Start with a clean baseline from billing export
A forecast built on the rough numbers in the console will inherit their noise. Pull the detailed billing export to BigQuery instead, which gives you SKU-level cost, usage, credits, and labels going back as far as you enabled it. Our guide to Cloud Billing reports and BigQuery billing export covers the setup. From that data, build a trailing twelve-month view of net spend by service and by project so seasonality and one-off spikes are visible. Strip out anything you know will not recur, such as a migration burst or a deleted environment, so the baseline reflects the true run rate rather than history.
Separate the drivers from the noise
Spend does not grow on its own; it grows because something underneath it grows. Identify the two or three drivers that actually move your bill: monthly active users, data volume in BigQuery, requests through Cloud Run, or nodes in your GKE clusters. Express the largest cost lines as a rate against their driver, for example dollars per million BigQuery queries or per active user. A driver-based forecast is far more defensible than a flat percentage, because when the business plan changes you can re-run the number instead of arguing about it.
Model committed use discounts and credits explicitly
Discounts are the part of a Google Cloud forecast that people most often get wrong. Sustained use discounts apply automatically and lower the effective rate on steady compute; if you are not sure how, read how sustained use discounts work. Committed use discounts are deliberate and need modeling: a one or three year commitment changes the effective rate on a slice of your spend for its whole term, so a forecast that ignores expiring or new commitments will be wrong by the discount amount. Our explainer on resource versus spend-based committed use discounts shows how each behaves. Track commitment start and end dates in the model so the forecast steps correctly when a term rolls off.
Want a forecast finance will trust?
Our Google Cloud cost audit builds a driver-based forecast on your billing export, models your commitments, and hands finance a number with the assumptions behind it. On the performance model, you pay only from realized savings. No savings, no fee.
Book a GCP cost audit →Build three scenarios, not one number
A single forecast figure invites false precision. Build a base case from current drivers and known plans, a low case that assumes the optimization work lands, and a high case that assumes growth runs ahead of plan. The spread between them tells finance how much the number can move and why, which is more useful than a point estimate that will be wrong by the second week. The low case is also where your optimization roadmap shows up as real dollars, which is how a forecast becomes a lever rather than a chore.
Track variance every month and correct
The step that separates a real forecast from a spreadsheet exercise is the monthly variance review. Each month, compare actual net spend to the forecast, by service and by project, and explain every line that missed by more than a set threshold. Variance is information: a project that ran hot points to either a bad assumption or unplanned growth, and either way you learn something that improves next month's number. Feed the explanation back into the driver rates and the forecast converges over a couple of quarters.
| Step | Input | Output |
|---|---|---|
| Baseline | BigQuery billing export | Clean run rate |
| Drivers | Business metrics | Rate per driver |
| Discounts | Commitment terms | Effective rate by month |
| Scenarios | Plan and roadmap | Low, base, high |
| Variance | Actuals vs forecast | Corrected assumptions |
Product names and discount mechanics above reflect Google Cloud as of May 2026. Verify current commitment terms and pricing in Google Cloud documentation before modeling, as they change.
The Google Cloud Cost Optimization Field Guide includes the forecast model template and the BigQuery queries behind the baseline. It is the downloadable companion to this article.
The short version
Build a clean baseline from billing export, express spend as rates against the drivers that actually move it, model your committed use discounts and credits by month, run a low, base, and high scenario, and review variance every month so the number self-corrects. A good forecast pairs with active controls; see how to set budgets and alerts to enforce it. When you would rather have the forecast and the savings plan built for you, that is what our Google Cloud cost optimization service delivers.