Return to User Guides
What is AI Insights?
AI Insights automatically transforms your enriched customer data into clear audience segments, named personas, and decision-ready strategic recommendations — without requiring any data science background. Once triggered, the system runs a four-phase pipeline: it cleans and prepares your data, groups customers into statistically validated segments, profiles each segment against the broader population, and uses Gemini 2.5 Pro to translate statistical findings into plain-language narratives and targeted next steps. The result is a structured, data-driven understanding of your customer base that your whole team can act on.Before You Begin
AI Insights works on audiences you’ve already matched and enriched through Huckle Match. Before running a summary, make sure you have at least one matched audience in your account.The richer your matched data, the more distinctive your segments will be. Lists with both email and address inputs — enabling Level 1 or Level 2 matches — produce the most differentiated clustering results. See Match Levels for details.
How It Works
Data Preparation & Feature Engineering
When a summary generation request is triggered, the system retrieves your audience data from secure cloud storage, cleanses it for consistency, and enriches it using feature dictionaries that translate raw fields into meaningful business attributes.Baseline statistics are computed across the full audience to establish a population-level reference point before segmentation begins.What this draws on: Demographics and Household attributes from your matched records, spanning the 350+ fields defined in the Data Dictionary.
Audience Segmentation Using K-Means Clustering
The system applies the K-Means clustering algorithm to identify natural groupings within your audience based on shared characteristics across demographics, behaviors, and spending patterns. The algorithm iteratively refines cluster centers until stable groupings emerge.Validation methods:
Outlier handling: An Isolation Forest model identifies customers with highly unusual patterns before clustering runs. This prevents edge cases from distorting segment definitions and enables them to be analyzed separately.This approach ensures segments are not arbitrary — they are statistically robust, clearly differentiated, and reflective of real customer behaviors. The output is comparable to what you see in Persona Clusters, but derived directly from your own audience data rather than the national Connex taxonomy.
| Method | Purpose |
|---|---|
| Elbow Method | Determines the number of segments that balances simplicity with explanatory power |
| Silhouette Score | Confirms customers are more similar within their segment than to others (target ≥ 0.5) |
Segment Profiling
Once segments are established, each one is profiled across demographics, behaviors, and spending patterns to build a clear picture of who each group is.Z-score analysis measures how far each segment’s attributes deviate from the overall population average. This surfaces the top 100 most distinctive traits per segment, highlighting what genuinely sets each group apart rather than just what’s common across everyone.What this draws on:
- Demographics — age, gender, marital status, income, home value, net worth
- Household — household composition, life stage, property characteristics
- Consumer Passions Index — 107 interest and affinity indices that surface lifestyle and behavioral differentiation
- MarketShare Demographics — 72 purchase behavior and transaction data fields for spending pattern analysis
AI-Powered Insight Generation
Gemini 2.5 Pro transforms statistical outputs into business-ready narratives through a structured two-layer approach.Layer 1 — Audience overview: Synthesizes overall audience patterns and benchmarks them against national averages to highlight what stands out about your customer base as a whole.Layer 2 — Segment narratives: Converts segment-level data into named personas, behavioral narratives, and targeted strategic recommendations.All AI outputs are schema-validated in real time to ensure consistency and prevent hallucinations.
AI-generated narratives and recommendations are grounded entirely in the statistical outputs from Phases 1–3. No assumptions are introduced beyond what the data supports.
End Result
A streamlined, data-driven understanding of your customer base — translated into clear segments, actionable insights, and decision-ready outputs your team can use immediately for creative strategy, media planning, prospect targeting, and retention programs.What You Can Do With AI Insights
| Output | How to use it |
|---|---|
| Named personas | Brief your creative and content teams with data-backed audience descriptions rather than assumptions |
| Behavioral narratives | Inform channel selection, offer strategy, and messaging tone for each segment |
| Strategic recommendations | Feed directly into campaign planning, product prioritization, or CRM segmentation logic |
| Segment definitions | Use the characteristics of your top segments as filter criteria in Audience Builder to find net-new prospects who match each group |
| Outlier profiles | Identify and separately analyze edge-case customers — churned users, anomalous high-value accounts, or one-time buyers |
What’s Next?
Persona Clusters
Explore national lifestyle and behavioral segments overlaid on your audience
Demographics
Dig into the demographic and financial profile of each segment
Audience Builder
Turn segment definitions into prospect lists for activation
Use Cases
See how AI Insights fits into end-to-end workflows