Explain the gap in your AI bill
Make the two figures comparable
Select one account and a completed UTC period. Keep currencies separate. Retain the exact source export, its collection time and coverage status. Compare the whole period rather than one displayed page of a report.
A gateway sees routed traffic. A provider account can include direct calls, console use and other workloads. Partial collection, missing usage chunks, stale price assumptions and cache categories can each create a difference. Team attribution still needs your own supporting records.
Follow a synthetic USD 70 difference
For the same example account and month, the gateway estimate is USD 1,180, the provider report is USD 1,200 and the invoice is USD 1,250. The first difference is USD 20. An example USD 15 direct job and USD 5 pricing correction explain it: 1,180 + 15 + 5 = 1,200.
The second difference is USD 50. In this synthetic invoice, USD 60 tax less USD 10 credit explains it: 1,200 + 60 − 10 = 1,250. Tax, credits and timing depend on the actual provider and agreement; these amounts are invented solely to show the arithmetic.
The three source totals remain separate. Adding them would count overlapping API usage several times. A zero residual is useful only when the explanations have source evidence; it is not proof that every charge is known.
Leave a record another person can check
Download the worksheet and keep one row for each bridge. Enter the currency, account label, UTC period, first and second source totals, signed adjustments, source reference and reviewer. Add your own formula in a spreadsheet if desired: second source minus first source minus explained adjustments is the unresolved residual.
Token Observe displays provider reports, imported invoice evidence and gateway estimates as distinct sources. The new cloud portal has no invoice CSV upload or automatic synchronization of existing imports. This guide is a manual review procedure, not a claim of automated invoice reconciliation or financial period close.
Where this argument meets an implementation
AI spend and subscriptions
What you were billed and what you were metered, recorded as two numbers and never added into one.
Spend controls
Hard USD ceilings, per-minute rate limits and a kill switch, all decided before the request leaves your network.
Those pages are one product's implementation of what this guide argues for, each listing what it does not do beside what it does.
How it worksDoes a missing source mean zero spend?
No. Record it as unknown and identify who can supply the evidence. A subtotal of known sources is not the whole estate’s spend.
Can the worksheet combine GBP, EUR and USD?
It retains the currency on every row. Compare only rows in the same currency and period; there is no exchange-rate conversion or blended total.
After reading it, the next step is running it on your own hardware.
Create a cloud account and workspace. No card or server installation. Model calls and Google BigQuery scans can incur separate provider charges.