Per-team, per-project, per-provider AI cost attribution

The black-box invoice problem

Anthropic's invoice arrives at month-end as "Claude API usage: £18,420". OpenAI's: "ChatGPT Teams: £6,200 + API: £8,100". You don't know which team drove the spend, which project consumed the most tokens, which engineers are over-investing in tokens vs delivering value. The natural FinOps response is to budget aggressively and complain. The accurate response is to know where the money went.

Receipt-level cost tracking

Every GenZAgents receipt logs runtime cost: model used + tokens in + tokens out → cost in USD/GBP. The dashboard rolls these up: per agent, per project, per team, per employee, per provider. You can filter "show me the top 10 cost drivers last month" — typically a handful of engineers running long autonomous agent loops will account for 60-70% of spend. That data is actionable.

The deliverable-level cost question

Beyond per-employee attribution, the high-leverage question is per-deliverable. "How much did the auth-service migration cost us in AI spend?" Without GenZAgents: impossible. With: receipts tagged with project=auth-service-migration sum to £342 across 47 receipts and 8 engineers. The CFO now has the per-deliverable cost basis they need to evaluate the AI investment.

Provider-level attribution for renegotiation

Receipt feed shows "62% of our AI spend is on Anthropic, 28% on OpenAI, 10% on Google". At renewal time, this is your negotiating leverage. You can credibly say "we spend £X with you, here's the contract size we'll commit to in exchange for Y% discount". Without the data, you're flying blind into vendor negotiations.

Anomaly detection on cost

A runaway agent loop can burn £500 in an hour. The GenZAgents anomaly detector flags cost spikes against a rolling 7-day baseline — alerts route to your Slack / PagerDuty in <5 minutes. The alert includes the offending agent / engineer / project so you can intervene. We've seen design-partner orgs catch four-figure cost spikes within hours of them starting.

Multi-currency + multi-provider rollups

For globally-distributed orgs: receipts in mixed providers and mixed currencies normalise to your reporting currency. Anthropic charges USD, GenZAgents records USD, your dashboard displays GBP at the rate published to receipts.daily_fx_rate. End-of-quarter rollups are clean and reportable.

Common questions

How accurate is the cost number on each receipt?▾

For Anthropic and OpenAI: within 1% of the actual invoice line item (we use published pricing × token counts). For Google: ~5% margin (their billing has more rounding). For self-hosted / Ollama: cost is null (no public pricing).

Can I export the cost data to my BI tool?▾

Yes — /v1/receipts has a CSV export endpoint that includes cost fields. Push it into Snowflake / BigQuery / your warehouse for joins with the rest of your spend data.

What about ChatGPT Plus seat licenses (not API)?▾

Per-seat licenses don't generate per-call cost — that's a fixed cost. The receipt for a ChatGPT browser-extension capture shows cost=null with provider=chatgpt-web. The dashboard rolls the fixed cost into the per-team line item via your manual configuration.

Does the cost include latency-driven over-spend?▾

No — it's strictly token-driven (input tokens × in-rate + output tokens × out-rate). Latency isn't billable. If you care about latency, see the receipt's runtime_ms field.