"How much will we spend on OCI next month?" It's a simple question that finance teams ask constantly — but answering it accurately requires more than just looking at last month's bill.
Why Forecasting Matters
Budget Accuracy
Finance teams need to forecast cloud costs for quarterly and annual budgets. Underestimating leads to budget overruns; overestimating leaves unused budget on the table.
Commitment Decisions
Deciding whether to purchase Universal Credits or reserved capacity requires understanding future spend. Committing based on current spend when you're about to scale up means missing potential savings.
Anomaly Context
Without a forecast, you can't distinguish between "costs are up because we're growing" and "costs are up because something is wrong."
Forecasting Techniques
1. Simple Trend Projection
Take the last 3-6 months of daily costs and project a linear trend forward. This is the simplest approach and works well for steady-state environments.
When it works: Mature, stable workloads with no planned changes
When it fails: Environments with planned migrations, seasonal traffic, or rapid growth
2. Weighted Moving Average
Give more weight to recent months than older months. A common weighting: 50% current month, 30% previous month, 20% month before.
This captures recent trends while smoothing out one-time spikes.
3. Seasonal Decomposition
If your workload has seasonal patterns (holiday traffic, end-of-quarter processing, fiscal year batch jobs), decompose your cost history into:
Trend: Long-term direction (growing, stable, declining)
Seasonal: Recurring patterns (weekly, monthly, quarterly)
Residual: Random variation
Forecast by projecting the trend and adding back the seasonal component for the target period.
4. Driver-Based Forecasting
Instead of forecasting costs directly, forecast the drivers of cost:
•Expected compute hours = expected users × average sessions × compute per session
•Expected storage = current storage + expected data growth
•Expected data transfer = expected API calls × average response size
Then apply pricing to derive cost estimates. This is more work but much more accurate for growing or changing environments.
Practical Tips
Use Multiple Time Horizons
7-day forecast: For operational awareness (are we on track this month?)
30-day forecast: For monthly budget tracking
90-day forecast: For quarterly planning and commitment decisions
Account for Known Changes
Simple trend analysis doesn't know about planned events. Overlay your forecast with:
•Planned infrastructure changes (migration from on-prem, new application launch)
•Seasonal events (Black Friday, fiscal year-end)
•Known price changes or contract renewals
Build a Confidence Range
Don't present a single number. Give a range:
Best case: Trend continues, optimizations are implemented
Expected: Trend continues, no changes
Worst case: Trend continues, plus buffer for unexpected growth
Review and Adjust
Compare actual costs to your forecast monthly. If you're consistently off by more than 10%, recalibrate your model.
Using OCIFinOps for Forecasting
OCIFinOps provides the historical data foundation for forecasting. Use the cost explorer to analyze trends by service and compartment, and the natural language query interface to quickly pull the data you need: "What was our average daily compute cost for each of the last 6 months?"
Accurate forecasting isn't about predicting the future perfectly — it's about being close enough to make informed decisions.