Reconciling outside AI spend without counting it twice
Compare provider bills with proxy records using matching periods, identities, and measurement bases.
A proxy ledger shows $40 of model usage and a connected vendor report shows $45. It is tempting to call the difference $5 of bypass spend and add the two totals to report $85 overall. Both conclusions can be wrong.
The records may overlap, cover different periods, or use different prices. Reconciliation starts by deciding which observations describe the same underlying work.
Match the measurement basis
Write down the currency, account, period, model coverage, and valuation basis of each source. Distinguish a list-price estimate from a vendor charge and an invoice from a usage snapshot. Check whether credits, discounts, and taxes are included before comparing amounts.
FOCUS provides a common specification for cost and usage data. Its value here is the discipline of preserving the meaning of a cost field rather than treating every monetary column as interchangeable. A connector still needs to document what its source actually supplies.
Reference: FOCUS: cost and usage specification
Work through a small example
Suppose an illustrative provider report contains $45 for a completed UTC day. A proxy record contains $40 for that same account and day at the same effective rates. If the proxy coverage is complete and the records refer to the same requests, the remaining $5 is a useful investigation target.
Now change one fact: the proxy window is the team's local calendar day. Requests near midnight move between periods. The difference can grow even if every call went through the proxy. Reconcile the timestamps before attributing the difference to a routing failure.
Keep the residual visible
Use an unresolved category when a difference could come from latency, missing usage, pricing, or unmapped identities. Record the latest successful import and allow completed periods to settle before escalating small differences. A stale connection should not produce the reassuring appearance of zero spend.
FlockTab's outside-spend views place supported vendor records beside its own meter. Imported costs are not governed by a tab merely because they appear next to it. A source can be connected and still have unassigned resources or an incomplete relationship to named Agents.
Investigate one mismatch end to end
Choose one account and one completed day. Find the matching proxy calls, the imported aggregate, and the relevant project mappings. Confirm the same model and pricing treatment. If the discrepancy survives those checks, inspect direct credentials and routes that might bypass the controlled path.
Document the outcome as a reporting gap, timing difference, price difference, or confirmed outside path. Then fix the cause that the evidence supports. A report becomes trustworthy when it can explain why totals disagree, not when it forces every difference into an alarming label.
Sources checked 23 September 2026. Numerical scenarios are illustrative unless explicitly identified as provider data. Vendor limits and interfaces can change; consult the linked documentation for current details.