Methodology
Unit costs
This page explains how CloudJaeger finds the cost of one customer or one feature.
Where it comes from
A customer's or a feature's cost comes from your cloud bills from AWS, Azure and Google Cloud.
Pinned by a test
Only net usage in US dollars is allocated: credits, refunds, tax and upfront purchases are not.
Pinned by a test
Bill rows in another currency whose conversion failed are not allocated, and each recompute says how many there are.
Pinned by a test
AI services billed on a cloud, such as Amazon Bedrock, Vertex AI and Azure OpenAI, are allocated like any other bill row.
Pinned by a test
Coming: AI providers you pay directly, such as OpenAI and Anthropic, and SaaS vendor bills are not yet split by customer or feature.
Allocated cost
Your allocation rules place each bill row on one customer, feature or team of each kind, and a row no rule claims is reported as unallocated.
Pinned by a test
A payer account and a member account whose costs it already bills are counted once.
Pinned by a test
A cost per unit for one customer or feature divides the cost allocated to it, not your whole bill.
Pinned by a test
A cost per unit with no customer or feature chosen divides your whole bill, and says so.
Pinned by a test
Large bills are summed in the database, so millions of bill rows are allocated without a limit on their number.
Pinned by a test
What is estimated
A bill row with no tag of its own can take the day's main tag value, but only when that value holds at least 60% of the day's tagged cost.
Pinned by a test
Cost placed that way is labelled estimated, month by month, beside the allocated total.
Pinned by a test
A month the figures do not cover day for day is labelled partial and is never compared with a whole month.
Pinned by a test
Coming: shared costs are not yet spread onto one customer's or one feature's cost per unit.
AI results
Cost per successful AI result is what the calls that reported an outcome cost, failed calls included, divided by the successful ones.
Pinned by a test
With no reported outcome there is no figure, never a zero.
Pinned by a test
A cheaper model's saving is not counted when your own quality scores put it more than two points below the current model.
Pinned by a test
Without enough quality scores, the saving is shown as quality unverified.
Pinned by a test
Coming: retry counts, because today a retried call counts as one more call.
Fresh or stale
The page states when the figures were last recomputed, and calls them stale when your bill has newer days or a rule changed since.
Pinned by a test
Coming: a daily refresh, because today the figures change when someone runs Recompute.
