How CloudZero matches AI usage to your bill

Each team, application, and person carries the AI spend their usage produced. CloudZero gets there by matching the AI usage your gateways and collectors send to the costs on your provider bills. If you connect provider bills only, your AI costs appear by the details on the bill itself, such as project, model, and user.

How the match works

Two kinds of data come together:

DataWhere it comes fromWhat it tells CloudZero
Provider billsYour Anthropic, OpenAI, and cloud provider connectionsWhat each model cost, and on which account or project
Usage telemetryYour AI gateways and collectorsWho used which model, and how much

CloudZero lines up the two on the details both sides share: the provider, the model, and the user. It then splits each part of the bill across the usage that matches it. The amounts come from your provider bill, and matching decides how that cost is shared across your teams, applications, and people.

How costs are shared across accounts

Bills also carry details that usage doesn't include, such as which account or project paid. Gateway telemetry can name the cloud account the gateway itself runs in, and that can be a different account from the one that paid for the model. CloudZero shares each model's cost across everyone who used that model, across all of your accounts for that provider. Each team's total reflects its share of that model's combined cost.

Send the detail CloudZero matches on

User and team details give each person and team their share of the bill. Set them on your gateway:

SourceWhat to setDetails
LiteLLMA user or team on each virtual keyConnecting to LiteLLM
BifrostA team or customer on each virtual keyConnecting to Bifrost
OpenTelemetryA user identifier on each requestConnecting with OpenTelemetry

Make each request traceable to your bill

These settings let you trace any request back to the account or project that paid for it, using your gateway's logs or your provider bill. That helps when you and your account manager look into a split that seems off.

SourceWhat to setDetails
LiteLLMA fixed model_info.id on each model entry in your LiteLLM config, with one provider account behind each entryConnecting to LiteLLM
BifrostProvider keys named after the account they billConnecting to Bifrost
OpenAIA dedicated OpenAI project for each gateway or applicationConnecting to OpenAI

Estimated and reconciled costs

AI Signals shows estimated costs within minutes of activity, so you can watch spend as it happens. Estimates use published rates.

Reconciled costs come from your provider billing and reflect your actual pricing. They appear across the platform alongside the rest of your cloud spend, including in Explorer. The cost figures in AI Signals' Overview and AI Explorer tabs are estimates. See Estimated vs. reconciled costs. Most billing arrives across the platform within about 24 hours, and some providers finalize recent costs later; Anthropic, for example, reconciles Enterprise billing for up to 30 days. Use reconciled costs for finance-grade reporting and chargeback.

Check how your spend is shared

This check uses a team or user Dimension that includes your AI costs. If you don't have one yet, set it up first.

  1. Pick one provider and a recent week that has finished billing.
  2. In Explorer, filter to that provider's AI costs for the week.
  3. Select Group By and choose your team or user Dimension.
  4. Check that each team carries the share of spend you expect.
  5. If a team's share looks off, contact your account manager and share the gateway's service name and the week you checked.

The estimated costs in AI Signals and your reconciled costs will differ, because estimates use published rates and your bill reflects your pricing. Compare which teams carry the most spend in each place.

ℹ️

Have questions or feedback? Reach out to your account manager.


Did this page help you?