AI spend chargeback: mapping every API call to a GL account automatically
Most organizations running AI across multiple teams face the same finance-side problem every month: one provider invoice arrives with one number, and finance has to allocate it across departments and GL accounts by hand — usually by asking engineering for an estimate, and usually getting one that's approximately right at best.
The fix isn't a better spreadsheet template. It's tagging the GL account and department at the moment each call is made, the same way feature-level attribution works, so the chargeback breakdown exists automatically by the time the billing period closes instead of being reconstructed after the fact.
Why the manual allocation process breaks down
Asking engineering to estimate departmental usage share after the invoice arrives has an obvious accuracy problem — nobody has a precise memory of which team's features drove which portion of last month's calls — but it also has a timing problem that compounds every cycle. The allocation has to be redone from scratch every billing period, by a person, which means the close takes three to five days longer than it would with a real breakdown, and that delay repeats indefinitely rather than getting fixed once.
The deeper issue is that the provider invoice was never built to carry this information. It reports model and token count, not department or GL account — there is no billing-layer setting that recovers organizational context the provider never captured in the first place, because chargeback isn't something a bill can express without help from the request that generated it.
How request-layer GL tagging solves it
Because Cognocient sits in the request path as a proxy, it can read chargeback context directly off each call's headers before forwarding it to the provider — the same mechanism that powers feature-level attribution, extended with headers finance actually needs. X-Cost-Department carries the business department; X-Cost-GL-Account carries the GL account code that matches your chart of accounts; X-Cost-Cost-Centre and X-Cost-Business-Unit add cost-centre and regional granularity for organizations that need it.
For teams that don't want to touch existing call sites, department and GL defaults can be set once on the API key itself in Settings, so every call made with that key automatically carries the right chargeback data with zero code changes — the easiest path to attributing legacy code that predates the tagging convention.
What finance actually gets at month-end
The Chargeback view shows a department summary table — spend, share of total, waste identified, and trend versus the prior cycle, per department — plus a GL account breakdown underneath it that maps spend directly to the chart of accounts finance already uses for journal entries. An unattributed-spend indicator surfaces the dollar value and percentage of calls that arrived without department headers, so tagging gaps are visible and fixable rather than silently folded into an “other” line.
The FOCUS 1.1 export includes the GL account code as a native column, so the file imports directly into SAP, Oracle Fusion, NetSuite, or QuickBooks without a remapping step — the GL account your team tagged at request time is the same code that lands in the ERP journal entry.
What this looks like in practice
Consider a hypothetical customer-success team whose ticket-resolution bot is tagged with X-Cost-Department: customer-success and X-Cost-GL-Account: 6420 on every call. By month-end, that spend already appears as its own line in the department summary table and its own row in the GL account breakdown — no reconciliation meeting, no engineering estimate required. Exporting the FOCUS CSV produces a file finance can post directly, with the GL account column already populated correctly.
The same mechanism extends to B2B chargeback: a SaaS company passing AI usage costs through to enterprise clients can use each client's account ID as the department value, producing a per-client line-item report suitable for inclusion in a client invoice or usage report. This is illustrative of how the mechanism behaves, not a specific customer's reported outcome.
Chargeback depends on attribution existing first — a GL account tag on an untagged call has nothing to attach to. See how request-layer attribution works for the foundation this builds on.