Your AI invoice says how much.
Not what you got.
Ingest joins your Anthropic and OpenAI token spend to the work it pays for: pull requests in GitHub, pipeline in Salesforce and HubSpot. One row per person, before and after AI, in your own data warehouse.
Then the CFO asks.
Token spend lives in Anthropic and OpenAI. The work lives in GitHub, Salesforce and HubSpot. So the answer is a pile of dashboards and spreadsheets, or nothing at all.
Token spend, joined to
the work it pays for.
Engineering and sales measured the same way: output per dollar, per person, against their own baseline.
One row per person.
What each person spends on AI, next to what they shipped or sold before and after.
ai_return_by_personSample data| Person | AI spend | Before → after AI |
|---|---|---|
| Maya R. | $412/mo |  PRs 18 → 29 |
| Ana L. | $96/mo |  Pipeline $180K → $236K |
| Theo K. | $74/mo |  Leads 41 → 63 |
| Jon P. | $388/mo | PRs 15 → 22 |
| Sam W. | $520/mo | PRs 21 → 21 |
Spend is read from Anthropic's and OpenAI's own cost reports, never estimated. Output is read from GitHub and your CRM.
And the spend
that returns nothing.
The same tables show where spend is not paying for itself.
Spend with no lift
People and keys whose spend rose while their output did not. Start the conversation with numbers instead of a hunch.
Model mix
Spend by model for each team, so routine work can move to a smaller model without guessing what it costs you in output.
Cache use
The share of input tokens read from cache, by API key. Low cache use on a repeated prompt is money left on the table.
All in your own warehouse.
The joined tables land in a warehouse or S3 bucket you own, so anyone can check how the number was made.
Connect the spend
Daily tokens and cost from Anthropic and OpenAI, by model, workspace and API key, with the members of each organization.
Connect the work
Pull requests and commits from GitHub. Leads, activities, opportunities and pipeline from Salesforce or HubSpot.
Match the people
Spend and output are joined on work email, with each API key tied to its owner. Spend no one can be matched to is shown, not dropped.
Measure the lift
Each person is compared with their own output before their AI spend began, and with teammates who have not started yet. Every number shows how many people stand behind it.
Questions
We pay for Claude or ChatGPT seats, not the API. Does this work?
For Claude seats, yes: Claude Enterprise reports activity per person through its Analytics API, and Ingest builds that connector when you ask for it. The key comes from your organization's primary owner. API usage from a Claude Console organization comes through the Anthropic (Admin API) connector. For any other plan that reports usage through an API, paste its docs link and Ingest builds the connector.
Is the lift proof that AI caused it?
No. It is a measured before-and-after with a comparison group, not a controlled experiment. Each figure shows how many people it rests on, so a number from three people reads like one.
What access does Ingest need?
A key for each source. For Anthropic that is an Admin API key, which only an admin of your Claude Console organization can create. Ingest uses it only to read usage, cost, members and keys.
Where does the data go?
Into a data warehouse or S3 bucket you own, refreshed on the schedule you choose. The tables are yours to query from any tool.
What does $1 of AI spend return?
The engineering view is three connectors: Anthropic, OpenAI and GitHub. The free trial runs up to 3 connectors for 14 days. See pricing