Problem Statement
The goal is to segment users into a small number of actionable groups based on behavioral patterns to support targeted marketing campaigns. The workflow must analyze event behavior, user attributes, and account age to generate clear segment labels, defining characteristics, messaging themes, and estimated segment sizes. The segmentation must emphasize behavioral signals and avoid creating overly small clusters.
Steps
Ingest user data
INGESTCollect user event logs, demographic attributes, and account age. Normalize and structure the data for behavioral feature extraction.
Retrieve historical behavior context
EXECUTELook up aggregate statistics such as average user activity, typical recency distributions, and common behavior patterns to guide segmentation thresholds and normalization.
Generate behavioral segments
EXECUTERun an LLM-assisted clustering analysis using behavioral features (action types, recency, frequency) as primary drivers. Produce 3–5 preliminary segments with rationale.
Validate segment actionability
EXECUTEConfirm each segment meets minimum size requirements, behavioral clarity, and practical marketing differentiation. Merge or adjust segments that fail constraints.
Generate segment labels and messaging
EXECUTEProduce descriptive segment names, key behavioral characteristics, messaging themes, and estimated segment sizes for marketing use.
Prompt Configurations
Behavioral segmentation prompt
You are a marketing analytics expert. Based on the provided user event data (actions, frequency, recency), demographic attributes, and account age, generate 3–5 actionable user segments. Requirements: prioritize behavioral patterns over demographics; ensure no segment is below 5% of users; describe the defining behaviors for each segment; provide strategic, insight-driven, marketing-ready language; do not invent data not provided. Output format: (1) segment label, (2) defining characteristics, (3) suggested messaging themes, (4) estimated percentage of users in each segment.
Example Inputs/Outputs
User dataset: 10,000 users. Behavior summary: 30% log in weekly and complete multiple actions; 25% log in monthly with low activity; 35% recently active but sporadic; 10% very old accounts with almost no recent behavior. Demographics included but not strongly correlated.
Segments: 1) High-Intent Power Users (30%): frequent sessions, high action volume; messaging: deepen engagement and upsell. 2) Low-Frequency Explorers (25%): infrequent logins, light usage; messaging: highlight core value props. 3) Sporadic Actives (35%): inconsistent but recent usage; messaging: reinforce habits with reminders. 4) Dormant Legacy Users (10%): very old accounts with minimal recency; messaging: reactivation campaigns. All segments meet the 5% minimum threshold and are behaviorally distinct.