About Segment Discovery
- Introduction
- Details about What Data Is Considered for Segment Discovery
- Analytic Settings That Affect Segment Discovery
Introduction
When an A/B test recommends a winning variant, segment discovery searches for segments of the target audience that respond well or poorly to a variant that is different from the recommended winner. A segment represents a group of people who share common characteristics. Impression rates, conversion rates, and segment size are analyzed to compare the performance of the winning variant to the remaining variants in an A/B test. By looking for common attributes among customer data, SAS Customer Intelligence 360 can discover a segment and find an alternate variant that performs better for that segment than the winning variant. The discovered segment is displayed on the Insights tab.
In order to discover a segment, SAS Customer Intelligence 360 tracks customer information, including conversions. The completion of a conversion is counted only once per visitor. For example, each of these scenarios counts as one conversion:
- A visitor views the variant and completes the conversion. The same visitor then leaves the site and returns later, without completing the conversion.
- A visitor views the variant but does not complete the conversion. Later, the same visitor returns to the site and completes the conversion.
- A visitor views the variant and completes the conversion. The visitor leaves the site. The same visitor returns later and completes the conversion again.
SAS Customer Intelligence 360 takes all of the historic impressions and conversion rates into account when comparing a variant against the winning variant. These comparisons are used to display statistics on the confidence between competing variants. By default, the confidence threshold for identifying a discovered segment is 90%. The variant that performs better for a discovered segment than the overall winning variant is identified as the winner for that particular discovered segment.
Details about What Data Is Considered for Segment Discovery
Overview of Data for Segment Discovery
Currently, only imported data, such as customer lists from SAS Marketing Automation or other sources, is used to discover segments. There is some data that is not analyzed when discovering segments. For example:
- a variable with too many missing values
- a character variable with too many possible values
- customer data that is discovered or collected by SAS Customer Intelligence 360, such as event attributes
Customer data is used for segmentation in these ways:
- Segment profiling is
done on a per-data item basis.
For example, assume that the structure of your customer data has columns for age and income. If a value for age is missing, the record is excluded from any calculations performed on age data. However, the record is still included in income calculations if an income value is present.
- Each data item from a descriptor is displayed as a criteria group for segments. These groups are identified when customer data is imported.
Import Settings for Segmentation
In order for customer data to be available when discovering segments, the data descriptor and segmentation properties must be set correctly before the data is imported into SAS Customer Intelligence 360. These settings are important when you want customer information (for example, gender) considered during segment discovery.
When the data is imported, these properties must be set in the data descriptor:
- segmentation must be
set to
true. - excludeFromAnalytics
must be set to
false. - segmentProfilingField
must be set to
true. This setting enables the user to be included on the User-Defined Criteria tab, which contains custom data items that you want to track. - tag values must include
DEMOGRAPHICSif you want to include the customer in segment profiling.If customer data is imported without this value, or customers visit your site anonymously, they can be used in a segment but cannot be part of segment profiling. For example, if your criteria are based on site activity, then the user could be part of the segment. However, the user is excluded from segment profiling for not having any demographic information.
Users are included in all other analytics services, regardless of associated attribute data.
- For a customer to appear on the Most Distinguishing Criteria tab for a discovered segment profile, there must be at least one attribute defined for the customer in the imported table. If there are no attributes defined for the customer's record, the customer cannot be considered when the set of highest differential visitors is compiled.
For more information about these properties, see Managing Customer Tables in SAS Customer Intelligence 360: Administration Guide.
Analytic Settings That Affect Segment Discovery
Navigate to 
Administration
Tasks to update these settings. The changes that you make to these settings
affect every user on your site.
Minimum Threshold to Discover Segments
After 10,000 users participate in an A/B test, those participants are categorized into different groups, or bins. Each group is analyzed for discovered segments. The number that you specify here determines the size of each group. After a discovered segment is found, it is displayed on the Insights tab. By default, the Minimum threshold to discover segments field is set to 1,000.
Set the minimum threshold so that SAS Customer Intelligence 360 discovers segments at a rate that you feel confident about. For example, if the minimum threshold is too low, the user interface displays many discovered segments, each made up of only a few members. As a result, you risk seeing evidence of segments that do not exist.
If the minimum threshold is too high, it can prevent discovered segments from being displayed. As a result, you might not see patterns that represent legitimate discovered segments.
Also consider the expected response rate for your A/B test. If you tend to get a high response rate from many participants, you can probably select a smaller number than if you expect a low response rate.
Try to set this threshold so that SAS Customer Intelligence 360 can discover three or four segments per test.
Calculate Sample Size
An A/B test cannot recommend a winner until the sample size is met. Segment discovery does not look for segments until a variant is declared a winner. As a result, if your sample size is so large that it cannot be met, segments cannot be discovered.
In addition, SAS Customer Intelligence 360 does not search for discoverable segments until a segment reaches 10,000 members. As a result, a small sample size can prevent segments from being discovered.