Video Central

The TVOD Sales Dataset provides transaction-level sales data for your Prime Video transactional video on demand (TVOD) business. Each record represents a completed transaction (purchase or rental). The data is delivered as an append-only changelog via the Slate Datasets API, giving you complete flexibility to build custom analytics and calculate metrics tailored to your business needs.

Key Benefits

Faster Insights — Sales data is batched multiple times per day, with most transactions delivered within ~9 hours of completion.
Cost and Sales Separation — Sales signals arrive in 4-hour batches; costing information (net_cogs) refreshes daily, so you get revenue signals as fast as possible.
Consistency — Standardized formatting across all territories in a single source.
Simplified Ingestion — Append-only changelog model with a simple upsert pattern for hands-off, automated ingestion.
Transaction-Level Granularity — Access individual order-level data to power custom analytics, title-level performance tracking, and internal system integrations.
Flexible Integration — Integrate TVOD sales data with your internal systems, data warehouses, and BI tools.

Feature	Slate Datasets API
Access Type	Programmatic (REST API)
Best For	Automated pipelines, enterprise reporting, custom analytics
Authentication	Login with Amazon (LWA) Security Profile
Data Format	CSV files (gzip compressed)

Getting Started

Prerequisites

Active Prime Video TVOD partnership
A login with Amazon (LWA) Security Profile
Client ID registered with your Content Account Manager (CAM)
An authorization code to request a token
A valid LWA authentication token for all API requests

Authentication Setup
To retrieve datasets you need to onboard to the Datasets API suite first. More details can be found here.

API Endpoints

Discovery Endpoints
Use these endpoints to look up your account and contract IDs programmatically:

Endpoint	Returns
GET /v2/accounts	List of Slate accounts you can access
GET /v2/accounts/{ACADIA_ID}	Business lines available (e.g., channels, transactions)
GET /v2/accounts/{ACADIA_ID}/transactions	List of TVOD contract IDs under your account
GET /v2/accounts/{ACADIA_ID}/transactions/{CONTRACT_ID}/datasets	Available datasets for a contract

Retrieving Dataset Files
Use this endpoint to retrieve links to transaction dataset files for your contract:

Request Parameters

Parameter	Description
ACADIA_ID	Your Slate account ID. Find it using GET /v2/accounts.
CONTRACT_ID	Your TVOD contract ID.
startDateTime	Set to the last time you pulled. Format: YYYY-MM-DDThh:mm:ssZ (UTC).
endDateTime	Set to current time. Format: YYYY-MM-DDThh:mm:ssZ (UTC).
limit	Maximum 1000 links per page.

This endpoint returns links to downloadable CSV files (gzip compressed) that contain the transaction logs — not the transactions directly.

Pagination
All responses are paginated. Use the following query parameters to navigate through results:

Parameter	Default	Description
limit	10	Number of documents returned per page. Maximum 1000.
offset	0	Number of pages to skip.

All paginated responses contain the following fields:

Field	Description
total	Total document count across all pages.
next	URL to the next page. Null if on the last page.

Data Model

Key Concepts

Concept	Description
Transaction	A completed order event (purchase or rental). Each transaction is identified by a unique order_item_id.
Changelog Model	Data is append-only. If a record’s attributes change (e.g., cost arrives), a new record is published with the same order_item_id but a newer last_update_time_utc.
Primary Key	order_item_id is the unique identifier for each transaction. Always deduplicate on this field.
Cost vs. Sales	Sales data arrives in 4-hour batches. Costing (net_cogs) refreshes daily, so records update with cost when available.

Data Freshness

Attribute	Target
Sales data delivery	Every 4 hours (batched)
Costing (net_cogs) refresh	Every 24 hours
End-to-end latency	~9 hours from transaction to data availability
Data retention	Maximum 2 years

Data Accuracy

Attribute	Details
Source of truth	Earnings and financial statements remain the final source of truth for payouts and costing.
Expected variance	Minor variance may exist due to date-time nuance and aggregation differences compared to financial/royalty reports.
Scope	Completed orders at the transaction grain for performance tracking. Not a replacement for financial or royalty reports.

This dataset is designed for faster, ongoing performance tracking. Expect minor variance compared to financial reports (e.g., Video ASIN Daily Level Summary) due to date-time nuance and aggregation differences. This is expected behavior, not a data quality issue.

Data Definitions

Core Fields
The following table describes all fields available in the transactions_event_log dataset:

Field Name	Type	Nullable	Description	Example
order_item_id	STRING	No	Unique identifier for each transaction. Primary key for deduplication.	ABC123XYZ
transaction_datetime_utc	TIMESTAMP	No	Transaction time in UTC.	2026-01-14T00:04:41.575
transaction_datetime_local	TIMESTAMP	No	Transaction time in local timezone.	2026-01-14T01:04:41.575
pv_title_id	STRING	No	Unique title identifier.	tt1234567
content_type	STRING	No	Type of content purchased.	Movie, TV Episode, TV Season
purchase_type	STRING	No	Purchase or rental indicator.	EST (purchase), VOD (rental)
content_quality	STRING	No	Video quality tier.	SD, HD, UHD
territory	STRING	No	Territory code.	US, GB, DE, JP, AU
device_class	STRING	Yes	Device category.	Fire TV, Mobile, Web
title_name	STRING	No	Title name.	The Great Adventure
currency	STRING	No	ISO 4217 currency code.	USD, EUR, JPY
vendor_sku	STRING	Yes	Partner-provided SKU.	WB-MOV-001
net_cogs	DECIMAL	Yes	Net cost of goods sold, excl. tax. Refreshes daily. May be NULL initially.	4.99
net_revenue	DECIMAL	No	Net revenue, excl. tax.	14.99
create_time_utc	TIMESTAMP	No	Record creation time in UTC.	2026-01-14T01:39:06.619
last_update_time_utc	TIMESTAMP	No	Record last update time. Used for deduplication logic.	2026-01-14T01:39:06.619

Field Notes
The following fields have important behavioral characteristics that partners should be aware of:

Field	Calculation / Logic	Notes
net_cogs	Refreshes daily	May initially appear as NULL or 0; updates within 24 hours when costing arrives.
last_update_time_utc	Latest timestamp wins	When multiple records exist for the same order_item_id, keep only the record with the latest last_update_time_utc.
territory	Single dataset for all territories	No per-territory files; filter by territory code as needed.
purchase_type	Fixed enum values	EST = Electronic Sell-Through (permanent purchase). VOD = time-limited rental.

Deduplication

Overview
The dataset uses a changelog model. When a record is updated (e.g., costing arrives), a new version is published with the same order_item_id and a newer last_update_time_utc. To maintain accurate data, always apply deduplication logic before writing records to your destination.

Deduplication Query
Use the following SQL pattern to deduplicate and retain only the latest version of each transaction:

Upsert Pattern
Use this pattern to maintain a local table with the latest state of every transaction:

ETL Pipeline

Overview
Follow this four-step process to build a reliable, automated ingestion pipeline for TVOD sales data.

Pipeline Steps

Initial Data Pull — Pull all files for your contract within the desired time range using the API endpoint. Download all files returned. Each file contains transaction records in CSV format (gzip compressed).
Deduplication — When multiple records exist for the same order_item_id while processing multiple days of data, keep only the latest last_update_time_utc record using the deduplication query in Section 6.2.
Upsert to Destination — Merge deduplicated records into your destination table using order_item_id as the key. See the upsert pattern in Section 6.3.
Incremental Processing — For ongoing data loads, set startDateTime to the last time you pulled and endDateTime to the current time. Process all returned files and upsert into your destination.

Set the following parameters for incremental runs:

Recommended Ingestion Cadence

Recommendation	Details
Recommended cadence	Every 4-6 hours to stay as current as possible.
Incremental processing strategy	Set startDateTime to the last retrieved timestamp and endDateTime to the current time. Process all returned files.
Daily/Weekly consumers	If you fetch daily or weekly, ensure you process all files for the full period to avoid missing records.
Recommended start time	For daily jobs, start at 1 AM UTC to capture prior day completions.

Sample Queries

Use these SQL patterns to get started with common analytics use cases. Replace [START_DATE], [END_DATE], and [X] with your desired values.

Top X Titles by Revenue Over a Period

Total Revenue by Purchase Type

Daily Sales Summary

Revenue by Territory

Quality Standards

Data Quality Targets

Quality Dimension	Target	Measurement
Completeness	>99% of transactions captured	Comparison against financial reports
Timeliness	~9 hours end-to-end latency	Time from transaction to data availability
Consistency	Single standardized format	All territories in one dataset

Known Limitations

Limitation	Description	Impact
Costing delay	net_cogs refreshes daily, not in real-time.	Records may initially show NULL or 0 cost; updates within 24 hours.
Data retention	Maximum 2 years of historical data.	Requests with timestamps older than 2 years will not return results.
Token expiration	LWA access tokens expire after 1 hour.	Must implement refresh token logic for uninterrupted access.
File-based delivery	API returns links to files, not direct data rows.	Requires a download step in your pipeline before processing.
Financial variance	Minor variance vs. financial/royalty reports.	Expected behavior due to date-time nuance; not a data quality issue.

TVOD Sales Dataset API