Predictive Analytics
Forecasting with two competing models and automatic selection, honest error metrics, risk scenario modelling, and change-impact simulation.
At a glance
| Route | /predictive-analytics |
| Group | Analyze |
| Page permission | intelligence |
What it is
Forecasting, risk scenario modelling, and action-impact analysis. Two competing time-series models run in parallel and the more accurate is selected automatically, with backtesting exposed so forecast quality is measurable rather than asserted.
Who it is for
Finance teams doing budget planning, FinOps leads setting targets, and engineering leaders planning capacity.
How it works
Two models, automatically selected
Both models run, accuracy is compared using MAPE and RMSE, and the better model is selected per workload.
The fallback is a trust decision
Where both models perform poorly, the system falls back to linear extrapolation rather than presenting an unreliable forecast confidently. A high error metric is itself useful information — it tells you the workload is not forecastable by these methods.
Change-impact simulation
Projects the effect of a proposed action on future spend — which is what allows an optimisation decision to be evaluated before it is taken.
Features
- Dual-model forecasting with automatic selection between Prophet and ARIMA
- Forecast horizons of 7, 14, 30, and 90 days
- 80% and 95% confidence intervals
- MAPE and RMSE error metrics exposed, not hidden
- Forecast backtesting against known history
- Linear extrapolation fallback when both models underperform
- Risk scenario modelling
- Action-impact analytics and change-impact simulation
- Seasonality decomposition including weekday and weekend separation
- Export of predictive analytics output
How to use it
Ensure sufficient spend history has been ingested
Forecast quality depends directly on history depth, and plan retention limits apply. Explorer's 30-day retention is not enough for a trustworthy 90-day forecast.
Generate a forecast at the horizon you need
30 or 90 days for budgeting. 7 or 14 for operational planning.
Check the backtest before relying on the number
A high MAPE means the workload is not forecastable by these methods — which is itself useful information, and better than a confident wrong number.
Read the confidence intervals, not the point estimate
Budget against the upper bound where overspend carries consequences.
Simulate proposed optimisations before executing
Use change-impact simulation to model the effect first.
Plan for growth and demand cases
Use risk scenarios, not just the base case.
Why it matters
Cloud budgeting is usually guesswork extrapolated from last month, which produces budgets that are either padded or breached.
Reliable forecasts with honest error metrics let finance set realistic budgets and engineering plan commitment purchases with confidence.
Exposing MAPE and RMSE is a deliberate trust decision: a forecast presented without its error is a number that cannot be responsibly used.
Connects to
- Computed within Intelligence
- Feeds budget guardrails in Policies and commitment sizing in Cost Center → Commitment Advisor
- Forecasting enablement is one of the eight FinOps Maturity dimensions
Kubernetes
Container cost allocation down to namespace, deployment, and team level, with rightsizing and idle-pod recommendations from measured container utilisation.
Cloud Playground
A scenario modelling environment for planning multi-cloud architectures, costed against live provider pricing with full provenance for every figure.