Three forecast models
Linear extrapolation: project month-to-date forward, trivial to compute, accurate for stable workloads only. Run-rate × growth: take prior-month spend × expected growth rate, works well for SaaS that scales with customers. Bottom-up: planned customers × per-customer infrastructure cost, most accurate but requires real unit economics. Mature forecasts blend models per project.
Confidence intervals
Single-point forecasts are easy to over-trust. Better forecasts report a band (e.g. $42K–$48K, 80% confidence) and decompose it: best case (high efficiency, low growth), expected case, downside (architectural change, faster growth, new launch). A finance team that commits to a single number with no band has under-priced uncertainty.
Who tovin.io is for
Frequently asked
How accurate should the cloud forecast be?
Within ±5% for stable SaaS workloads, ±10% for fast-growing or recently-relaunched products. Beyond ±15% the forecast is more wish than model — that's a signal to switch methods, not to forecast harder.
What's the most common forecasting mistake?
Forecasting the headline number without per-project forecasts underneath. The total is the sum of the parts; without the parts, you can't explain variance when actuals miss.