Forecasting and Budgeting

A cloud forecast combines the usage trend with known business and engineering changes; budgets turn the forecast into team-level targets with alerts on actual and forecasted spend.

Finance must tell leadership what technology will cost next quarter and next year. In the cloud that number moves with every release, so forecasting is a shared job between finance and engineering.

flowchart LR
  H["History: 6–12 months of cost"] --> B["Baseline trend"]
  K["Known changes: launches, migrations, commitments"] --> F["Forecast"]
  B --> F
  F --> BU["Budgets per team or account"]
  BU --> AL["Alerts: actual and forecast"]
  AL --> V["Variance review"]
  V --> F
  %% step H: Start from allocated history, ideally amortized cost so purchases do not distort it.
  %% step B: A trend model (provider forecast or a simple regression) gives the baseline.
  %% step K: Engineering and product add what the trend cannot know: launches, migrations, decommissions.
  %% step F: The forecast is baseline plus drivers, with a range rather than one number.
  %% step BU: Budgets set the target per team, account or cost centre.
  %% step V: Each month, explain the variance and update the forecast — that is how accuracy improves.

Forecasting methods

Method	How	When
Trend-based	Extend the last months, e.g. provider forecasts (Cost Explorer, Azure, GCP)	Stable workloads, short horizon
Driver-based	Cost per unit × forecast units (orders, users, GB)	Growth that follows business volume
Bottom-up	Each team estimates its own plans	Annual planning, big changes

Most teams combine them: trend for the baseline, drivers for growth, bottom-up input for known projects.

Forecast accuracy

Measure variance = (actual − forecast) ÷ forecast. Many practices aim for a monthly variance within a few percent at Run maturity, and accept wider ranges at Crawl. Track it per team: the goal is that teams explain their variance, not that the number is perfect.

Budgets

Set budgets per owner (team, account, cost centre), not just one company total.
Alert at several points: e.g. 80% of actual, 100% of forecasted spend.
AWS Budgets, Azure budgets and GCP budgets can notify email, chat or topics, and some can trigger budget actions (for example applying a restrictive policy) — use those carefully, mostly for sandboxes.
A budget is a target for a conversation, not a hard stop for production.

Traps: forecasting only from the total bill (one big change hides everything else); budgets nobody owns; and using billed cost so a large upfront purchase looks like a huge overspend in one month.