What your provider dashboard hides
Provider consoles report spend by API key. That tells you which credential paid, not which team decided to. Here is the gap, and what closing it takes.
Every AI provider gives you a usage dashboard. None of them give you an accounting system. The distinction matters more than it sounds.
Keys are not people
A provider console aggregates by API key, because that is the only unit it can see. Your organisation does not spend money per key — it spends per person, per team, and per decision. One shared key behind a service means one line item and no way to attribute it.
The result is a number that is accurate and useless at the same time: you know exactly what you paid and nothing about who committed you to paying it.
Three questions a console cannot answer
- Who is shipping value per token? Token counts rank the verbose, not the effective. The useful ratio is tokens to accepted output.
- What share of every context window was paid waste? Re-sending context you did not need is invisible in a usage graph and material on an invoice.
- Which spend never needed a model at all? Some fraction of any workload is a task a short script would have handled.
What closing it takes
Instrumentation at the call site, not the billing boundary. Tag every request with the person and the team behind it, record the input/output split, and keep the trace long enough to compare cost against whatever you count as a shipped result.
That is a modest amount of engineering, and it changes the conversation with finance from what did we spend to what did we get.