Which business application?:

Google Actions Center — the partner portal Google uses to onboard

businesses into AI Mode / Gemini as action providers. Distinct from

Google Ads. Angi is publishing a nightly Merchants feed so Google can

AI-match our ~1.4M indexable pros to Google Business Profiles and

surface Angi as a home-services action provider in AI Mode. Reference:

https://developers.google.com/actions-center/verticals/reservations/bl/reference/feeds/merchants-feed

What object in the destination?:

A "merchant" record — one record per business (pro). Google's Merchants

feed schema requires:

- merchant_id (opaque, URL-safe, immutable identifier issued by the

  partner)

- name (business name)

- category (required; constant "home_service" applied for all Angi

  merchants)

- geo.address (required; country="US", locality=city, region=state,

  postal_code=zip)

Optional Google fields we omit for v1: telephone, geo lat/long,

action_link, url, trust signals. Business-profile hydration (URLs,

ratings) happens via a separate Pro Lookup API keyed on merchant_id

— not via the feed itself.

Volume: ~1.4M merchant records per nightly snapshot.

Delivery format: JSON, gzip-compressed, sharded (~200 MB per shard,

~1M records per shard). Each shard carries a metadata envelope with

processing_instruction, shard_number, total_shards, nonce, and

generation_timestamp.

How will records be identified?:

Primary identifier is merchant_id — Angi's canonical business UUID,

issued by us and used opaquely by Google (also serves as the join key

for our companion Pro Lookup API). Dedupe at UUID grain: at most one

merchant record per UUID. When a single UUID appears as both an ADS

and a LEADS row in our source data, the LEADS / canonical row wins.

What sync behaviour?:

Full-snapshot replacement per run (Google's processing_instruction =

PROCESS_AS_COMPLETE). Google has deprecated incremental processing for

Merchants feeds, so incremental / upsert semantics are not applicable;

each nightly delivery fully supersedes the previous snapshot.

Cadence: nightly (one full snapshot per 24 hours).

Delivery mechanism (critical for connector design):

- Transport: SFTP to partnerupload.google.com:19321

- Auth: SSH key-based (partner uploads public key to Actions Center;

  private key held on the sender side)

- Files: gzip-compressed JSON shards, ≤200 MB compressed each,

  numbered as shard_number / total_shards

- Environment separation: upload username differs between production

  and sandbox (Google issues separate feed-user accounts per env)

- Metadata envelope wrapping the merchant records must include

  processing_instruction, shard_number, total_shards, nonce, and

  generation_timestamp

Sandbox testing: Google provides a sandbox upload user for validation

against processing reports before production go-live; the connector

should support toggling between sandbox and production usernames per

environment.

Additional details: