segmentation

name	segmentation
description	Split users, customers, orders, or other entities into meaningful subgroups (RFM, behavioral, value-based, lifecycle) and analyze each subgroup separately. Use when an aggregate metric likely hides important variation across cohorts.

When to segment

Aggregate metrics often hide divergent behavior across subgroups. Segment before drawing conclusions when:

• The user asks "who are our best/worst X" or "how does X vary across groups"
• A trend looks flat but you suspect offsetting movements (e.g. SMB churn masked by enterprise growth)
• You're sizing or prioritizing — different segments deserve different strategies

Common segmentation types

• RFM (Recency, Frequency, Monetary) — retention and marketing on transactional businesses
• Behavioral — by feature usage, lifecycle stage, engagement level
• Value-based — ARR tier, deal size, plan
• Demographic / firmographic — for B2C / B2B targeting
• Acquisition — by channel, campaign, signup cohort

How to evaluate a segmentation

1. Size — each segment should be material (rule of thumb: ≥5% of base, or large enough for statistical confidence)
2. Distinctness — segments should differ on the metric you care about, not just on the segmenting variable
3. Stability — membership shouldn't churn so fast the segment becomes meaningless

Pitfalls

• Tiny segments — outliers dominate; either merge or exclude
• Overlapping segments — if you'll do per-segment math, prefer mutually exclusive splits
• Backward causation — segmenting by an outcome (e.g. "high-value customers") and then explaining the outcome with that segmentation is circular