Data specification

9 min

Amazon S3 export is being decommissioned on November 30, 2026. See the GCS partitions and migrate using the Authentication APIAuthentication API.

Use the Experimentation Events Export when you need to analyze experiment data outside of Optimizely. For example, you might analyze the data in a data warehouse, business intelligence tool, or custom statistical pipeline. The Experimentation Events Export gives you access to the following data:

Decision events (also known as impressions) – Events Optimizely Experimentation fires when it assigns a visitor to an experiment-and-variation pair.

Decision events in Optimizely Feature Experimentation.

Decision events in Optimizely Web Experimentation.

Conversion events (also known as events) – Events Optimizely Experimentation fires when a visitor converts. For example, a click, page view, or purchase.

Conversion events in Optimizely Feature Experimentation.

Conversion events in Optimizely Web Experimentation.

Your export contains the same decision and conversion event data Optimizely uses to render the Experimentation Results page. Your analysis therefore aligns with what the Results page displays. For more, see How Optimizely Experimentation counts conversions.

Event content

Your event data includes the following fields, among others:

Event timestamp.

Visitor ID.

Event (entity) ID and event name.

User attributes.

Event tags.

Additional visitor and event metadata.

Server process timestamp.

As of October 9, 2023, Optimizely no longer generates the session_id field on decision events, or the session_id and experiments fields on conversion events.

For more, see the Experimentation Analytics October 2023 release notesExperimentation Analytics October 2023 release notes.

For the complete field reference, see the Schema section.

Event data transformation

Optimizely preprocesses the data in the Experimentation Events Export to ensure accuracy, consistency, and usability. These transformations run on every export, so consumers downstream see the cleaned data, not the raw client events. The export applies the following transformations:

Remove duplicate events – Deduplicates events based on uuid.

Populate missing event names – Sets event_name to entity_id if event_name is blank.

Filter out incomplete conversions – Removes a conversion unless entity_id, revenue, or value is populated, or event_type is client_activation.

Replace oversized values – Substitutes event_name or experiment_id if the original value exceeds size limits.

Access the export data

The data is stored in Parquet-formatted file parts. The number of file parts depends on your total event volume. Use parquet-tool to inspect file contents and schemas.

Your data is available in Amazon S3 and Google Cloud Storage (GCS). The format and schema are identical. Only the storage location and credentials differ. See the Authentication APIAuthentication API for how to obtain credentials for each.

The S3 data is encrypted using AWS-KMS. To access it, you need one of the following clients:

AWS CLI version 1.11.108 or later

AWS SDKs released after May 2016

The GCS data is encrypted at rest by Google Cloud Storage. To access it, use the Google Cloud CLI (gcloud storage) or a Google Cloud Storage client library.

AWS S3 partitions

Your event data is exported in the following S3 partitions:

Decisions – s3://optimizely-events-data/v1/account_id=ACCOUNT_ID/type=decisions/date=YYYY-MM-DD/experiment=EXPERIMENT_ID

Conversions – s3://optimizely-events-data/v1/account_id=ACCOUNT_ID/type=events/date=YYYY-MM-DD/event=EVENT_NAME

Legend

optimizely-events-data – S3 bucket name.

account_id – Your unique account identifier.

date – The data creation date.

experiment_id – Unique experiment identifier (used for the decisions partition).

event_name – Event (entity) identifier (used for the conversions partition).

GCS partitions

Your event data is exported to a per-account GCS bucket in the following partitions:

Decisions – gs://optimizely-e3-ACCOUNT_ID/v1/account_id=ACCOUNT_ID/type=decisions/date=YYYY-MM-DD/experiment=EXPERIMENT_ID

Conversions – gs://optimizely-e3-ACCOUNT_ID/v1/account_id=ACCOUNT_ID/type=events/date=YYYY-MM-DD/event=EVENT_NAME

Legend

optimizely-e3-ACCOUNT_ID – GCS bucket name (one bucket per account).

account_id – Your unique account identifier.

date – The data creation date.

experiment_id – Unique experiment identifier (used for the decisions partition).

event_name – Event (entity) identifier (used for the conversions partition).

Status file

The daily partition files are ready for import when the _SUCCESS file is available.

Amazon S3

Decisions – s3://optimizely-events-data/v1/account_id=ACCOUNT_ID/type=decisions/date=YYYY-MM-DD/_SUCCESS

Conversions (Events) – s3://optimizely-events-data/v1/account_id=ACCOUNT_ID/type=events/date=YYYY-MM-DD/_SUCCESS

Google Cloud Storage

Decisions – gs://optimizely-e3-ACCOUNT_ID/v1/account_id=ACCOUNT_ID/type=decisions/date=YYYY-MM-DD/_SUCCESS

Conversions (Events) – gs://optimizely-e3-ACCOUNT_ID/v1/account_id=ACCOUNT_ID/type=events/date=YYYY-MM-DD/_SUCCESS

Poll for this file. When it is present, import the daily partition data.

File structure

The Experimentation Events Export has the following file structure:

Decisions – Visitor bucketing information for experiments. Bucketing is the process of assigning a visitor to one of the variations in an experiment, based on their identifier and the experiment's traffic allocation.

Experiment

Events (conversions) – All tracked events, such as a click, a hover, or any other configured event.

Event

Schema

Decisions

Each row represents a single decision event.

Field	Type	Description
`uuid`	string (GUID)	Event UUID generated by the client. Used to deduplicate events.
`timestamp`	long (time millis)	Event timestamp – UTC, UNIX epoch milliseconds.
`process_timestamp`	long (time millis)	Server process timestamp – UTC, UNIX epoch milliseconds. Indicates when the server processed the event.
`visitor_id`	string	The user or visitor identifier set by the client.
`session_id`	string	Unique session ID. As of October 9, 2023, Optimizely no longer generates `session_id`.
`account_id`	string	Account identifier.
`campaign_id`	string	Campaign ID.
`experiment_id`	string	Experiment ID. Set to `null` if the user is evaluated for the experiment and fails the audience condition.
`variation_id`	string	Unique variation ID. Set to `null` if the user is evaluated for the experiment and fails the audience condition.
`attributes`	array\	An array of user attributes (also known as segments).
`user_ip`	string	User IP address.
`user_agent`	string	User-agent.
`referrer`	string	Client referrer (the page from which the event was sent).
`is_holdback`	boolean	Indicates whether traffic allocation excluded the visitor from the experiment. When `true`, the visitor was excluded.
`revision`	string	Client snippet revision.
`client_engine`	string	Client engine string (for example, `js` or `node-sdk`).
`client_version`	string	Client version.
`activation_id`	string	Available only for Web Experimentation products, including Performance Edge. Used to calculate bounce and exit metrics.
`anonymize_ip`	boolean	When set to `true`, the last octet of the IP address is removed before storage. When `false`, the full IP address is stored. Most relevant if your implementation runs in a web browser or mobile app under data-privacy regulations. This flag operates independently of the IP-anonymization setting in Account and Project settings. When `true`, IP-based filtering does not work, because anonymization runs before IP filters apply.

As of October 9, 2023, Optimizely no longer generates session_id. See the Experimentation Analytics October 2023 release notesExperimentation Analytics October 2023 release notes.

Conversions

Each row represents a single conversion event.

Field	Type	Description
`uuid`	string (GUID)	Event UUID generated by the client. Used to deduplicate events.
`timestamp`	long (time millis)	Event timestamp – UTC, UNIX epoch milliseconds.
`process_timestamp`	long (time millis)	Server process timestamp – UTC, UNIX epoch milliseconds. Indicates when the server processed the event.
`visitor_id`	string	User identifier set by the client.
`session_id`	string	Unique session ID. As of October 9, 2023, Optimizely no longer generates `session_id`.
`account_id`	string	Unique account identifier.
`experiments`	array\	An array of the campaigns, experiments, and variations the event is attributed to. As of October 9, 2023, Optimizely no longer generates the `experiments` field.
`properties`	map \	Key-value pairs that define properties of the event. The value must be a string.
`entity_id`	string	Event entity identifier.
`attributes`	array\	An array of user attributes (also known as segments).
`user_ip`	string	User IP address.
`user_agent`	string	User-agent.
`referrer`	string	Client referrer (the page from which the event was sent).
`event_type`	string	Event type – one of `click`, `pageview`, `custom`, or `client_activation`.
`event_name`	string	Human-readable event name from the client, or `multi-event` when multiple events are batched.
`revenue`	long	Revenue (in cents).
`project_id`	string	(Optional) Pass `project_id` only when you are using the Optimizely Recommendations product.
`value`	double	The value used to compute value or numeric metrics.
`quantity`	long	Quantity metric value.
`tags`	map\	Key-value pairs of event tags.
`revision`	string	Client snippet revision.
`client_engine`	string	Client engine string (for example, `node-sdk`).
`client_version`	string	Client version.
`activation_id`	string	Available only for Web Experimentation products, including Performance Edge. Used to calculate bounce and exit metrics.
`anonymize_ip`	boolean	When set to `true`, the last octet of the IP address is removed before storage. When `false`, the full IP address is stored. Most relevant if your implementation runs in a web browser or mobile app under data-privacy regulations. This flag operates independently of the IP-anonymization setting in Account and Project settings. When `true`, IP-based filtering does not work, because anonymization runs before IP filters apply.

As of October 9, 2023, Optimizely no longer generates session_id or experiments. See the Experimentation Analytics October 2023 release notesExperimentation Analytics October 2023 release notes.