Segmentation Analysis

Purpose

Identify and characterize meaningful groups in a population in ways that support different decisions for each group.

When to use

Use this skill when:

• asked "who are our best customers?" or "how do we group our users?"
• building targeting lists for different campaigns, offers, or interventions
• conducting RFM (Recency, Frequency, Monetary) analysis
• evaluating whether different customer segments have different needs or behaviors
• deciding which users to prioritize for retention, upsell, or onboarding attention

When not to use

Do not use this skill when:

• the goal is to understand why a KPI changed (use root-cause-analysis)
• the goal is to predict future behavior for individual entities (requires a predictive model)
• segments are already defined and documented - just apply them rather than re-deriving

Required thinking discipline

• A segment is only useful if you would do something different for each group. If the action is the same for all groups, the segmentation adds no value.
• Define the decision use case before defining the segments. Segments derived backward from a decision question are more actionable than segments derived from data patterns alone.
• Validate segment stability. A segmentation scheme that reclassifies 60% of customers month over month is operationally unusable.
• Name segments by behavior, not by rank. "High-value retained buyers" is more actionable than "Tier 1."
• Evidence constraint: Every conclusion must cite specific data — a number, a rate, a segment, or a timeframe. Do not speculate without evidential basis. If data is insufficient, state what is missing rather than asserting an unsupported inference.

Workflow

1. Define the decision use case: Why are we segmenting? What action will differ by segment? (Examples: different email cadences, different onboarding tracks, different discount thresholds, different retention interventions.)

2. Choose the segmentation basis:

   • Behavioral: what the entity does (purchase frequency, feature usage, engagement level, product mix)
   • RFM: Recency (when did they last transact?), Frequency (how often?), Monetary (how much value?)
   • Attribute-based: firmographic (industry, size, geography), demographic, plan type
   • Lifecycle stage: new, active, at-risk, churned, reactivated

3. Define the time window and entity scope: Which time period defines the behavioral signals? Are we segmenting all users, or a specific cohort?

4. Assign entities to segments: Define the boundaries clearly - use explicit thresholds rather than vague labels. Example: "Frequent" = 4+ purchases in the last 90 days, not "more than average."

5. Characterize each segment: For each group, describe:

   • Size (count and % of population)
   • Key behavioral metrics (average revenue, frequency, recency, product usage)
   • Distinguishing characteristics compared to other segments
   • What makes this group different and why it matters

6. Validate segment stability: Check whether segment membership is consistent across two or three comparable time periods. High churn between segments indicates the boundaries are noisy or the time window is too short.

7. Map segments to actions: State explicitly what you would do differently for each segment. If the action is the same, merge the segments.

Output format

• Decision use case and segmentation basis
• Segment definitions (explicit thresholds or rules)
• Segment profiles table (size, key metrics, distinguishing behaviors)
• Stability assessment
• Action mapping (what is different for each segment)
• Recommended monitoring cadence

Good example

Segmentation basis: RFM for ecommerce customer base, last 12 months.

Segment	Definition	Size	Avg Revenue	Avg Orders	Action
Champions	R <= 30d, F >= 6, M top 20%	8%	$890	9.2	Loyalty reward, early access
At-Risk High Value	R 61-120d, F >= 4, M top 30%	6%	$640	5.1	Win-back campaign, discount offer
New Customers	R <= 30d, F = 1	22%	$110	1.0	Onboarding sequence, second purchase nudge
Dormant	R > 180d	31%	$95	1.4	Low-cost reactivation or suppress

Stability check: 84% of Champions retained segment membership month over month. Segment boundaries are stable.

Bad example

Our customers can be grouped into High, Medium, and Low.

Why this is bad:

• no definition of what "high," "medium," and "low" mean behaviorally
• no thresholds - who exactly falls into each group?
• no stability check
• no action mapping - what would you do differently for each group?
• the segmentation is not connected to any decision

Practical notes

• RFM is the most robust entry point for ecommerce segmentation because all three dimensions are directly actionable.
• For B2B SaaS, account health score (combining login recency, feature adoption, support tickets) often works better than transactional RFM.
• Segment count should reflect the number of distinct actions you can actually execute - if you can only run two campaigns, four segments add no value over two.
• When presenting segments to a business team, name them descriptively (Champions, At-Risk, New) rather than with cluster IDs (Segment 0, Segment 1) - the names signal the action.

Optional variants

• RFM (ecommerce): use recency, frequency, and monetary value with explicit quantile thresholds. Standard output is a 3x3 or 5x5 grid.
• Product engagement tiers (SaaS): use feature adoption depth, session frequency, and time-to-value as the behavioral signals. Tie segments to onboarding and expansion playbooks.
• Account health scoring (B2B): combine product usage, stakeholder engagement, support volume, and contract signals into a composite score. Map to customer success intervention triggers.
• Lifecycle segmentation: new, active, at-risk, churned, reactivated. Useful when retention is the primary objective and behavioral signals are less granular.