Cloud Cost Forecasting

Cloud Cost Forecasting

Forecasting projects your future cloud spend based on historical patterns. Xplorr uses two models — linear trend and seasonal — to give you a 60-day projection at the account, service, and organization level.

How it works

Linear regression model

Xplorr runs an ordinary least squares (OLS) regression on your past 90 days of daily spend. This produces a trend line that projects forward 60 days.

The model gives you:

• Projected end-of-month (EOM) spend — what your current month will likely total
• Trend direction — increasing, decreasing, or flat
• Trend magnitude — percentage change month-over-month (e.g., +12% or -5%)
• Confidence interval — upper and lower bounds based on variance in your historical data

Seasonal forecast model

Some workloads have predictable weekly patterns — lower spend on weekends, spikes on batch-processing days. The seasonal model captures day-of-week patterns and adjusts the forecast accordingly.

The seasonal model activates automatically when Xplorr detects statistically significant day-of-week variation (F-test, p < 0.05). You don’t need to configure anything.

What you see in the console

Open Forecasting from the left sidebar in the console. The main view shows:

• Month-to-date (MTD) spend — actual spend so far this month
• Projected EOM spend — where the month is heading
• Trend indicator — percentage change vs previous month
• 60-day forecast chart — daily projected spend with confidence bands

Example reading:

MTD: $5,400 (day 15 of 30) Projected EOM: $8,200 Trend: +12% vs last month Confidence: $7,600 – $8,800 (90% interval)

Budget integration

Forecasting ties directly into your budgets. When the projected EOM spend exceeds a budget threshold, Xplorr sends a projected overrun alert — even if your actual MTD spend hasn’t hit the threshold yet.

This gives you early warning. If your budget is $8,000 and the forecast projects $8,200 by day 15, you have 15 days to take action instead of finding out on day 28.

Projected overrun alerts are sent via the same channels as budget alerts (email and Slack). They include:

• Current MTD spend
• Projected EOM spend
• Budget amount and the projected overage
• Top 3 services driving the increase

Forecast accuracy

The forecast gets more accurate as the month progresses:

• Day 1–7: Wide confidence interval, +/- 25%
• Day 8–15: Narrowing, +/- 15%
• Day 16–25: Reasonably tight, +/- 8%
• Day 26–30: Very accurate, +/- 3%

Xplorr tracks forecast accuracy over time. You can see the historical accuracy on the Forecasting page under Accuracy History — it shows the projected vs actual EOM spend for each of the past 6 months.

Step-by-step: using forecasts for planning

1. Check the forecast weekly. Every Monday, open the Forecasting page and review the projected EOM for your organization. Compare it to your budget.

2. Drill into services. If the overall trend is up, click into the service-level breakdown to see which service is driving the increase. Common culprits: compute auto-scaling, storage growth, or data transfer spikes.

3. Combine with recommendations. If the forecast shows you’ll exceed budget, check the Recommendations page for quick wins — idle resources or rightsizing opportunities that can bend the trend down.

4. Set a projected overrun alert. If you don’t have one, go to Budgets, edit your budget, and enable the Forecast Alert toggle. Choose how many days in advance you want to be alerted.

5. Review accuracy at month end. After the month closes, compare the forecast to actuals. If accuracy is consistently off, check whether you have volatile workloads (like spot instances or batch jobs) that the linear model can’t capture well.

Real-world example

A team’s March 2026 forecast on day 15:

Metric	Value
MTD spend	$5,400
Projected EOM	$8,200
Budget	$7,500
Trend	+12% vs February
Top driver	EC2 (+18%), RDS (+8%), S3 (+3%)

The projected overrun alert fired on day 12, giving the team 18 days to respond. They:

• Rightsized 3 EC2 instances (saved $320/month)
• Deleted unused EBS snapshots (saved $85/month)
• Scheduled non-critical batch jobs to use spot instances (saved $200/month)

Actual EOM spend: $7,380 — under budget.

Common mistakes

• Relying on the forecast in the first week of the month. The confidence interval is wide early on. Use it directionally, not as a precise number.
• Not accounting for one-time events. A data migration or load test will spike costs temporarily. The linear model treats it as a trend shift. Wait for the spike to pass and the forecast will self-correct.
• Ignoring the seasonal model. If your batch jobs run on weekdays only, the seasonal model gives a more accurate forecast than the linear model. Check which model Xplorr is using on the Forecasting page — it’s shown in the chart legend.

FAQ

Can I forecast at the tag level? Not yet. Forecasting currently works at the organization, account, and service level. Tag-level forecasting is on the roadmap.

How much history does the model need? At least 30 days of data for the linear model. The seasonal model needs 60 days. If you just connected your account, forecasts will be available after the minimum history period.

Does the forecast account for reserved instances and savings plans? Yes. The forecast is based on your actual billed costs, which already reflect RI and SP discounts.

Can I export the forecast data? Yes. Click Export on the Forecasting page to download a CSV with daily projected values, confidence intervals, and the underlying trend parameters.

Key takeaways

• The forecast is most useful as an early warning system — pair it with budget alerts to catch overruns before they happen.
• Drill into service-level forecasts to identify which workloads are driving cost increases.
• Forecast accuracy improves as the month progresses — make decisive actions after day 15 when the confidence interval tightens.
• Review forecast vs actuals monthly to calibrate your expectations and improve planning.