cost-analysis

Cost analysis

Turn the scanner's collected billing data into a cost breakdown and a forward forecast. The scanner collects up to 12 months of Cost Explorer history per account, which is what makes both possible.

Tools

• get_total_cost — total spend across accounts, with per-account totals and AWS Organizations context.
• get_cost_by_service — spend broken down by AWS service.
• get_cost_trends — month-by-month history (the basis for the forecast).
• get_cost_comparison — compare spend between two periods or scans.
• get_organizations_cost_breakdown — use this in Organizations/Control Tower environments when asked which account costs the most: it separates the master/payer account's direct costs from member-account usage, avoiding the common trap where the payer account appears to dominate.

Workflow

1. Scope

Confirm which account(s) and time range the user means. Use list_scans / get_scan_summary to identify the account. State the account and the period the data covers before presenting numbers.

2. Breakdown

• Headline: get_total_cost for the overall figure and per-account split.
• By service: get_cost_by_service — report the top services by spend and what share each represents.
• Multi-account / Organizations: prefer get_organizations_cost_breakdown for "which account costs the most", and explain the master-vs-member distinction it returns.
• Change over time: get_cost_comparison when the user wants period-over-period.

Always state the currency (the tools return it) and the exact months covered.

3. Trend and forecast

• Call get_cost_trends for the monthly series (trends, month_count).
• Describe the trend: rising, falling, flat, seasonal, or spiky.
• Forecast from the history — the server does not forecast, so you compute it and show your method:
  • Use a simple, stated method: a 3-month trailing average for a stable series, or a linear projection of the last several months for a clear trend.
  • Project the next 1–3 months and give a range, not false precision.
  • State every assumption (method, months used, that it extrapolates past usage and ignores planned changes, Reserved Instance/Savings Plan expiry, and one-off charges).
  • Call out anything that undermines the forecast: too few months of data (say so and widen the range), a large one-off spike, or an obvious step change. If there are fewer than three months of history, do not forecast — report the trend only and say why.