Multi-Cloud Billing Export Analyzer

About the Multi-Cloud Billing Export Analyzer

Load the cost export your cloud already writes — an AWS Cost and Usage Report (the legacy CUR or CUR 2.0 from Data Exports), a FOCUS export from any provider, an Azure cost export or a Google Cloud billing export — and see where the money goes by service, account, region, usage type, resource and tag, day by day.

The analyzer flags unusual days and new services, finds idle and waste (unused reservations and savings plans, public IP addresses nothing uses, NAT gateways and load balancers without traffic, gp2 volumes, older instance generations, development machines running at weekends), measures untagged spend and forecasts the month. Switch between billed cost (as invoiced), effective cost (commitments amortized) and list cost at any time.

Files are read in your browser by a background worker — CSV, CSV.GZ, ZIP, Parquet, JSON lines and Excel — so reports of millions of lines in many parts work, and nothing is uploaded. Without a pass the report is a free preview: every total and count, the forecast, the first rows of each list and the chart under a mark; the full report and the Excel, PDF and CSV files unlock with a Premium pass.

Common uses

• Finding out which team, service or account drove this month’s bill up, and on which day it started.
• A monthly cost review for a startup or agency without a paid FinOps platform, from the files the cloud already exports.
• Cleaning up waste before buying Savings Plans or reservations: idle addresses, NAT gateways, load balancers and unused commitments.
• Measuring tag coverage before turning on chargeback, and listing the untagged resources to fix first.

Which exports it reads

• AWS: the legacy Cost and Usage Report (lineItem/UnblendedCost columns, CSV, CSV.GZ or ZIP), CUR 2.0 and FOCUS 1.0 from Data Exports (line_item_unblended_cost columns with the resource_tags and product maps, CSV.GZ or .snappy.parquet parts), the Athena layout of the legacy CUR, and Cost Explorer’s CSV download (a day per row and a column per service). Column names: CUR 2.0 table dictionary, FOCUS 1.0 with AWS columns, file names and parts.
• FOCUS from any provider: BilledCost, EffectiveCost, ListCost, ChargeCategory, PricingCategory, CommitmentDiscountStatus, Tags (FOCUS specification).
• Azure: cost details of Enterprise Agreement, Microsoft Customer Agreement and pay-as-you-go accounts (CostInBillingCurrency, Cost or PreTaxCost, MeterCategory, ResourceGroup, ChargeType, PricingModel, Tags; dates month first) (the fields).
• Google Cloud: the BigQuery billing export flattened to CSV (service.description, sku.description, project.id, cost, credits, cost_type, labels) or exported as JSON lines or Parquet with its nested fields (export structure).
• Any other cost table with a date and a cost column: the columns are guessed and can be chosen by hand.

How the figures are worked out

• Billed cost is the invoice amount: AWS unblended cost, FOCUS BilledCost, Azure CostInBillingCurrency, Google Cloud cost plus the credits in each row (Google’s own invoice query adds SUM(cost) + SUM(credits.amount)).
• Effective cost spreads commitments over the usage they cover. For the AWS CUR, Savings Plan covered usage counts at SavingsPlanEffectiveCost, reserved usage at the reservation’s effective cost, upfront fees and negations at 0, and recurring fees only for their unused part (Savings Plans in the CUR); FOCUS gives EffectiveCost directly.
• Unusual days are higher (or lower) than the median of the 14 days before by more than three scaled median absolute deviations (1.4826 × MAD), by at least a quarter of that median and by at least 1 % of a typical day of the whole bill. A service whose first charge comes after the first week is listed as new.
• The forecast adds, for each day left, the median of the last 7 complete days without one-time charges (purchases, tax and credits that are not posted every day); the range uses the 20th and 80th percentiles of the last 14 days and the trend their Theil–Sen slope. A last day that stops early (exports refresh during the day) counts as still to come.
• Untagged spend is the share of usage with a resource id that carries no tag (or not the chosen tag); tax, credits, purchases, support and usage without a resource id cannot be tagged and are left out.

What the idle and waste findings look for

• Unused commitments: the unused part of reservations and savings plans (AWS RIFee and SavingsPlanRecurringFee lines, Azure UnusedReservation and UnusedSavingsPlan, FOCUS CommitmentDiscountStatus = Unused). Beside the totals, reservations and savings plans used is their utilisation: the covered usage at its amortized cost ÷ (that plus the unused part).
• Idle public IP addresses: AWS PublicIPv4:IdleAddress and ElasticIP:IdleAddress usage, Google Cloud “Static Ip Charge”.
• NAT gateways and load balancers without traffic: hourly charges with under 1 GB a month processed, or under 0.01 load balancer capacity units an hour.
• gp2 volumes: AWS documents gp3 at a 20 % lower price per GiB, so the estimated saving is a fifth of their cost.
• Non-production resources running around the clock: tagged dev, test or staging, charged every day including weekends; needed 12 hours on weekdays, they would run 60 of 168 hours, about 64 % less.
• Older instance generations (AWS t2, m4, c4, r4 and earlier, Google N1, Azure Av2 and Dv2), a large data-transfer share and steady on-demand compute that a commitment could cover, with a link to the AWS Savings Plans and Reserved Instances planner.

Limitations

• An export shows cost, not use: CPU, memory and request metrics are not in it, so right-sizing needs your monitoring data. Every finding says what to check before acting.
• Exports of different currencies are analysed separately, never added together. Days are UTC days.
• Very large files take time on a phone; a laptop reads a few hundred thousand lines a second. Up to 300,000 resources are kept one by one; any more are added together in one row.
• Brotli-compressed Parquet files are not read (AWS, Azure and BigQuery write Snappy, gzip or Zstandard, which are). Encrypted or password-protected ZIP files must be unpacked first.

Privacy

On this device: nothing is sent

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot.