Deliveries land on Mondays. The cadence and file mix depend on your tier. You receive an email notification when your delivery is ready.
Cadence by tier
Starter deliveries land every other Monday (biweekly). Growth and Scale deliveries land every Monday.
File formats
Starter customers receive a JSONL file. Growth and Scale customers receive the JSONL file plus four Parquet files (videos, people, brands, products), a JSON Schema file describing the current schema version, and a manifest JSON blob.
JSONL is newline-delimited JSON, one complete VideoObject per line. It is the canonical shape: every field in every record, no joins needed. Load it into Snowflake with COPY INTO against a JSON file format, into BigQuery with bq load using NEWLINE_DELIMITED_JSON, or parse it with any streaming library in a scripting language.
Parquet is columnar and much faster to load into a warehouse. The four files are joinable on video_id: videos.parquet holds one row per video, people.parquet holds one row per video-person pairing, and brands.parquet and products.parquet work the same way. Load Parquet into Snowflake with COPY INTO using the Parquet file format, into BigQuery with LOAD DATA, or query it directly in DuckDB with read_parquet() without loading at all.
Manifest
Every Growth and Scale delivery includes a manifest JSON blob with delivery_id, period_start, period_end, schema_version, record counts, unique people and brand and product counts, the list of industries covered, file sizes in bytes, and a quality scorecard (completeness percentage, average entities per video, and error count). The manifest is the single source of truth for what a delivery contains. Read it first if you are scripting ingestion.
Retention
File retention depends on your tier. Starter: 90 days. Growth: 180 days. Scale: 365 days. Retention is measured from the delivery date and is frozen at the tier in effect when the delivery was made. After expiration, the file is removed from storage but the delivery record stays visible in the portal. If you need longer retention, archive deliveries to your own storage after downloading.
Cloud Routing
Scale customers with Cloud Routing configured get files pushed directly to one dedicated endpoint (S3, GCS, Azure Blob, or a webhook) the moment the delivery is ready. See the Cloud Routing article for setup details.
Notifications
When a delivery lands, an email goes out to every team member who has the download role. The portal activity feed records a data.delivery_received event, visible to admins from the activity view. Make sure at least one member has download access before your first delivery week.
Detailed reference
What you can expect
- Monday delivery across every tier
- JSONL newline-delimited records, one VideoObject per line
- Parquet columnar files joinable on video_id (Growth and Scale)
- Manifest JSON blob with delivery_id, counts, and quality scorecard
- Delivery notification email and portal activity feed entry
- File retention: Starter 90 days, Growth 180 days, Scale 365 days
Limits
- Mid-week delivery drops (all tiers land Monday mornings)
- Parquet files on Starter (JSONL only)
- Custom cadences on Starter or Growth (Scale can request daily)
- File retention beyond 90 days (archive to your own storage)
What to do
- Read the manifest first when scripting ingestion, it tells you exactly what is in the delivery
- Use Parquet for warehouse loading, use JSONL for streaming or ad hoc parsing
- Archive deliveries to your own storage if you need history beyond 90 days
- Ensure at least one team member has the download role so notification emails get delivered
If something looks off
If a Monday passes without a delivery notification, check the portal activity feed first for a data.delivery_received event. If the event is missing, contact [email protected] with your organization name and the expected delivery date. If the event is present but no email arrived, verify that at least one team member has the download role.
Important context
If Monday falls on a US federal holiday the delivery still ships. Schema validation runs on every JSONL record before the file leaves our systems, so malformed records never reach you.
Last updated 2026-04-15