AI Discovered Audiences & Behavioral Trends
While traditional query builders require marketers to hypothesize segments in advance, Keystone's AI Discovered Audiences engine autonomously uncovers hidden behavioral clusters and revenue opportunities across your customer base.
1. Unsupervised Audience Clustering
Navigate to Segments -> Discovered Audiences (/segments/discovered) to inspect machine-learning proposed cohorts.
Keystone runs automated unsupervised clustering models (DBSCAN and K-Means) across customer feature vectors: - Session timing patterns (e.g. night-time shoppers vs weekday morning browsers). - Category browsing diversity and price elasticity. - Channel responsiveness (SMS clickers vs email openers vs organic returnees). - Trajectory velocity towards purchase.
2. Proposed Cohort Cards
Each proposed audience card details: - Cohort Label: Auto-generated descriptive title (e.g., Weekend Flash Spenders, High-Velocity Discount Searchers, Lapsed Premium Subscribers). - Cluster Size: Total profiles identified in this behavioral group. - Conversion Likelihood: Predicted conversion probability compared against the baseline average. - Distinctive Trait Vector: Key features that set this group apart from the general population. - Action Button: One-click conversion from an AI proposal into an official, active Dynamic Segment.
3. Behavioral Trends & Anomaly Detection
Navigate to Analytics -> Behavioral Trends (/analytics/behavioral-trends) to observe macro shifts in customer behavior over time.
What the Trend Analyzer Monitors:
- Sudden Drop in Category Engagement: Detects when interest in a primary product category drops significantly week-over-week.
- Shift in Device Adoption: Tracks migration from desktop web to mobile devices to inform UI optimization.
- Spike in Cart Abandonment: Highlights checkout step friction immediately, rather than waiting for month-end financial reviews.