CONNECTIONS / Databricks / Databricks to OpenAI Offline CAPI
Your Databricks lakehouse, wired to OpenAI's Offline Conversions API.
Signals reads the Delta table through a Unity Catalog grant and reports downstream revenue as OpenAI's conversion measurement comes online, hashed on the way out and governed like any other lakehouse job.
- ● databricks → openai.capi :: live
- > read main.sales.pos_ledger rows=51,915
- hash sha256(email,phone) consent=filtered
- route openai deliver
- ✓ delivered · 40,494 matched
WHAT THIS ENABLES
Databricks to OpenAI Offline CAPI
Databricks to OpenAI Offline CAPI: A notebook job reads the offline and in-store sales in your Databricks Delta tables, and Signals reports downstream revenue as OpenAI's conversion measurement comes online, hashed and consent-checked each time the model rebuilds.
- Signals reports downstream revenue as OpenAI's conversion measurement comes online, reading your Databricks Delta tables each time the model rebuilds.
- A revenue feed ready the moment OpenAI's offline measurement expands, fed by data engineers modeling in Databricks notebooks.
The click, the conversion, and the credit.
The conversion happens off OpenAI, away from any pixel. Here is how Databricks closes the loop.
Built for the teams that own the number.
Data engineers modeling in Databricks notebooks who need a revenue feed ready the moment OpenAI's offline measurement expands.
From kickoff to verified events.
-
Connect
A Unity Catalog-scoped service principal on the single Delta table in play to feed OpenAI's Offline Conversions API, sized for the emerging OpenAI ad surface.
-
Map
A notebook job reads your Databricks Delta tables, and its columns map to OpenAI's Offline Conversions API in the visual mapper, hashed and checked for the emerging OpenAI ad surface.
-
Deliver
Each time the model rebuilds, Signals reports downstream revenue as OpenAI's conversion measurement comes online, reads only the rows the model rebuilt, with the emerging OpenAI ad surface watched in the debugger.
What changes when the CSV goes away.
| Capability | Manual CSV upload | Datahash |
|---|---|---|
| Reporting | Offline sales sit in a separate export, reconciled by hand. | Revenue counted in the same Offline CAPI reporting as web conversions. |
| Deduplication | A re-uploaded file risks counting the same conversion twice. | Deduped delivery, so a resent record never counts as a second conversion. |
| Match visibility | Match quality is a guess until the numbers look off. | Match rate reported per upload, tied to the source event set that produced it. |
| Effort and latency | An analyst exports and uploads on a manual cadence. | Server-side and automatic the moment Databricks records the event. |
What Offline CAPI actually receives.
- ● signals :: event payload
- > POST /offline_conversions source=databricks.main.sales.pos_ledger
- event_name "Purchase" · event_time 1784660956
- em sha256 "2b9d64…" · value 25107 · currency AED
- ✓ accepted · row_status delivered
Adjacent moves on the same stack.
from Databricks
Databricks to Google OCI
Databricks to Google Offline Conversions: A notebook job reads the offline and in-store…
Learn morefrom Databricks
Databricks to Google Store Sales
Databricks to Google Store Sales: A notebook job reads the offline and in-store sales in…
Learn more
same use case
ActiveCampaign to OpenAI Offline CAPI
ActiveCampaign to OpenAI Offline CAPI: Signals reads your ActiveCampaign contacts and deal…
Learn moregeneric
OpenAI Offline Conversions API
Datahash uploads offline and in-store conversions to OpenAI with hashed identifiers, so…
Learn moreAsked on almost every call.
How often does the Databricks OpenAI Offline CAPI sync run?
Cadence follows the job that rebuilds the table: a nightly model gives OpenAI a nightly feed, an hourly one an hourly feed. Because a run picks up only the rows the model rebuilt, a revenue feed ready the moment OpenAI's offline measurement expands keeps pace on the emerging OpenAI ad surface and job-cluster time stays modest.
What match rate should a Databricks-sourced OpenAI Offline CAPI batch expect?
Identifier quality in the Delta table drives it, not the cluster size. Rows carrying a live hashed email or phone land; rows with neither will not, and no rebuild changes that. The opening run's feedback from OpenAI's Offline Conversions API flags the gap for early advertisers on OpenAI inventory measuring revenue from day one tracking the emerging OpenAI ad surface.
NEXT STEP