Streaming segmentation
Streaming segmentation on Adobe Experience Platform allows customers to do segmentation in near real-time while focusing on data richness. With streaming segmentation, segment qualification now happens as streaming data lands into Platform, alleviating the need to schedule and run segmentation jobs. With this capability, most segment rules can now be evaluated as the data is passed into Platform, meaning segment membership will be kept up-to-date without running scheduled segmentation jobs.
Streaming segmentation query types query-types
A query will be automatically evaluated with streaming segmentation if it meets any of the following criteria:
A segment definition will not be enabled for streaming segmentation in the following scenarios:
- The segment definition includes Adobe Audience Manager (AAM) segments or traits.
- The segment definition includes multiple entities (multi-entity queries).
- The segment definition includes a combination of a single event and an
inSegment
event.- However, if the segment definition contained in the
inSegment
event is profile only, the segment definition will be enabled for streaming segmentation.
- However, if the segment definition contained in the
- The segment definition uses “Ignore year” as part of its time constraints.
Please note the following guidelines apply when doing streaming segmentation:
- The lookback window is limited to one day.
- A strict time-ordering condition must exist between the events.
- Queries with at least one negated event are supported. However, the entire event cannot be a negation.
If a segment definition is modified so it no longer meets the criteria for streaming segmentation, the segment definition will automatically switch from “Streaming” to “Batch”.
Additionally, segment unqualification, similarly to segment qualification, happens in real-time. As a result, if an audience no longer qualifies for a segment, it will be immediately unqualified. For example, if the segment definition asks for “All users who bought red shoes in the last three hours”, after three hours, all the profiles that initially qualified for the segment definition will be unqualified.
Streaming segmentation segment definition details
After creating a streaming-enabled segment, you can view details of that segment.
Specifically, the Total qualified metric is displayed, which shows the total number of qualified audiences, based on batch and streaming evaluations for this segment.
Underneath is a line graph that shows the number of new audiences that were updated in the last 24 hours using the streaming evaluation method. The dropdown can be adjusted to show the last 24 hours, last week, or last 30 days. The New audience updated metric is based on the change in audience size during the selected time range, as evaluated by streaming segmentation. This metric does not include the total qualified audience from the daily segment batch evaluation.
Additional information about the last segment evaluation can be found by selecting the information bubble next to Total qualified.
For more information about segment definitions, please read the previous section on segment definition details.
Next steps
This user guide explains how streaming-enabled segment definitions work on Adobe Experience Platform and how to monitor streaming-enabled segments.
To learn more about using the Adobe Experience Platform user interface, please read the Segmentation user guide.
Appendix
The following section lists frequently asked questions regarding streaming segmentation:
Does streaming segmentation “unqualification” also happen in real time?
For most instances, streaming segmentation unqualification happens in real-time. However, streaming segments that use segments of segments do not unqualify in real-time, instead unqualifying after 24 hours.
What data does streaming segmentation work on?
Streaming segmentation works on all data that was ingested using a streaming source. Segments ingested using a batch-based source will be evaluated nightly, even if it qualifies for streaming segmentation. Events streamed into the system with a timestamp older than 24 hours will be processed in the subsequent batch job.
How are segments defined as batch or streaming segmentation?
A segment definition is defined as batch, streaming, or edge segmentation based on a combination of query type and event history duration. A list of which segments will be evaluated as a streaming segment definition can be found in the streaming segmentation query types section.
Please note that if a segment definition contains both an inSegment
expression and a direct single-event chain, it cannot qualify for streaming segmentation. If you want to have this segment definition qualify for streaming segmentation, you should make the direct single-event chain its own segment.
Why does the number of “total qualified” segments keep increasing while the number under “Last X days” remains at zero within the segment definition details section?
The number of total qualified segments is drawn from the daily segmentation job, which includes audiences that qualify for both batch and streaming segments. This value is shown for both batch and streaming segments.
The number under the “Last X days” only includes audiences that are qualified in streaming segmentation, and only increases if you have streamed data into the system and it counts toward that streaming definition. This value is only shown for streaming segments. As a result, this value may display as 0 for batch segments.
As a result, if you see that the number under “Last X days” is zero, and the line graph is also reporting zero, you have not streamed any profiles into the system that would qualify for that segment.
How long does it take for a segment definition to be available?
It takes up to one hour for a segment definition to be available.
Are there any limitations to the data being streamed in?
In order for streamed data to be used in streaming segmentation, there must be spacing between the events streamed in. If too many events are streamed in within the same second, Platform will treat these events as bot-generated data, and they will be discarded. As best practice, you should have at least five seconds between event data in order to ensure the data is properly used.