Methodology

Segmentation systems structured like PersonaLive have existed for a long time. The core advance is combining new behavioral data sources with demographics—and the models that bind them—so households can be differentiated on more than census variables, with rich segment profiles that do not depend on traditional survey panels alone.

Sources

PersonaLive is built from four primary source families (aligned with how the platform overview describes classification):

1. Social media — Who people follow and what they engage with (including geolocated and non-geolocated posts and following lists)
2. Spend data — Credit and debit card patterns
3. Location data — Places visited via mobile signals
4. Demographic data — Census and individual household attributes (age, income, marital status, and similar)

Older documentation sometimes emphasized social and mobile first; spend is a first-class input in the current product, including Strategist-plan spend tools.

Clustering

Variables are selected for coverage, predictive value, and low redundancy. Combinations of variables and weights are tested to find clustering that performs best across predictive evaluations.

Clustering is performed at the census block group level, with household-level data used to enhance and re-allocate household classification into the final 80 segments and 17 groups.

Evaluation

PersonaLive iterations are evaluated against real-world outcome data across industries and use cases. Datasets used in evaluation have included (public and private):

• Sales
• Subscription membership
• E-commerce
• Mobility / visitation behavior
• Insurance
• Retail sales
• Additional proprietary outcome sets

Evaluations compare PersonaLive iterations to each other and to competing segmentation systems. Statistical measures such as cluster inertia are also used to assess how well a model separates records.

Related

• Dataset Structure — Segment and group codes
• Overview — Platform workflow and tools
• FAQ — Common questions about segments and indexes