Cost Forecasting#
Changed in version 1.0.0: The forecasting engine was completely rebuilt in v1.0.0. The new system automatically adapts to each resource’s spending pattern — whether costs are flat, growing, seasonal, or volatile — and switches to a better model whenever behaviour changes. Accuracy is significantly improved, especially for Reserved Instance and Committed Use workloads.
Reply CMP provides a 30-day cost forecast for every resource in your inventory. The forecasting model learns continuously from incoming billing data — no manual intervention required.
How it Works#
The forecasting engine runs 9 parallel strategies per resource simultaneously — flat/stable, smoothed average, trending, step-change, weekly seasonal, monthly seasonal, and three ARIMA variants. Each strategy is continuously scored against actual costs using an exponential moving average of its prediction error. The engine automatically switches to whichever strategy is performing best for that specific resource.
This means:
Resources with steady flat costs get a simple stable model
Resources with weekly cycles get a weekly seasonal model
Resources with growth trends get a trending model
Resources that change behaviour (e.g. after a resize) automatically switch to the better strategy within days
Predictions update on every billing sync and also whenever reservation data is refreshed — keeping forecasts accurate for Reserved Instance, Savings Plan, and Committed Use Discount workloads.
When a resource’s spending behaviour changes — for example after a resize or migration — the model adjusts within a few days without any manual action.
Warm-Up Period#
A new resource needs at least 7 days of billing history before a forecast is produced. Until then, the resource appears in the UI with “Insufficient data”.
Reading Forecasts in the UI#
Forecasts appear in two places:
FinOps → Analyze: the cost trend chart shows the 30-day forecast as a dashed line. Hover over any future day to see the expected value.
FinOps → Assess: the “Forecasted Spend” summary card shows the 30-day total for the selected scope (org, group, or connection).
Accuracy#
Forecast accuracy depends on how consistent the resource’s cost pattern is:
Pattern |
Typical accuracy |
|---|---|
Stable / reserved |
Very high — further improved in v1.0.0 |
Predictably seasonal |
High after a few cycles of data |
Steadily growing |
Good |
Highly variable / spiky |
Lower — extreme spikes may be smoothed out |
Note
Forecasts are advisory. Use them for budget planning and anomaly baselining, not as financial commitments. Large one-off events (surprise charges, credits, billing adjustments) will temporarily reduce accuracy until the model adapts.
FAQ#
Why does my forecast look flat? There is not yet enough variation in the data to identify a trend or seasonal pattern. As more billing data arrives the forecast will become more expressive.
The forecast dropped to zero for a resource — why? If a resource had days with no cost (e.g., it was turned off, or billing data arrived late), those gaps are filled with near-zero values. If a resource was decommissioned, the forecast will naturally decay toward zero.
Can I see a forecast beyond 30 days? The platform supports longer forecast horizons. The default view currently shows 30 days.