How it works
What happens to a dollar of AI spend.
It arrives on a provider bill with nothing attached to it. It leaves with an owner, a purpose, an outcome and a price per outcome.
Every fold below is one step of that, in the order it happens.
What you run
What you get back
Start where your data already lives
Connect a provider, gateway or existing export.
No agent on employee devices. No prompt or response content.
Provider connection
Read-only, one vendor.
Start with one provider's usage and billing data, reconciled against what we compute.
Gateway or telemetry
One credential, everything behind it.
See the providers, services and agents already flowing through it.
Export or app inventory
Your directory's app list, or an expense export.
Find AI vendors and spend that sit outside engineering systems.
Step two
Four actions. One record.
Each action reads and writes the same reconciled record, so the loop compounds instead of restarting.
Spend Explorer
$27,144 30 days · 5 sources
| Source | Spend | Share | Vendor bill |
|---|---|---|---|
| Amazon Bedrock | $24,288 | 89% | $24,288 |
| Anthropic | $2,568 | 9% | $2,569 |
| OpenAI | $287 | 1% | $287 |
Step three · Outcome Packs
Somebody has to say what counts as success.
The ledger divides any cost by any number. Which number, and what it excludes, is the one decision a machine cannot make.
A cost tool answers
What did the support AI cost last month?
$21,787
An Outcome Pack answers
What does one resolved case cost?
$0.69
The system your team already closes work in.
The helpdesk's closed cases, joined to the agent runs that worked them. Nothing new to instrument, and no change to how the team works.
Support, worked through
34,399
cases recorded in the window
The record
One reconciled record. Every view agrees with it.
Regroup it by provider, model, agent, team, workflow or outcome. The dimensions change; the total does not.
Connect one source and watch it happen.
Read-only, and the first reconciled total in minutes.