Every average in your analytics is a lie of composition. "Average session: 7 minutes" might describe no actual user — half your base pops in for 40 seconds, a devoted core lives in the product for 20 minutes. Segmentation is how you stop optimizing for a fictional average user and start serving the real ones.

Here are the three segmentation models that matter for SaaS, when to use each, and how to keep the whole thing actionable instead of academic.

Why segment at all

Three payoffs, all concrete:

Diagnosis. A churn rate of 4% means little; discovering it's 1.5% for teams and 9% for solo users (illustrative) tells you exactly where the leak is.
Prioritization. Features, onboarding paths, and support effort can target the segments that drive revenue rather than the loudest voices.
Message fit. Lifecycle emails, pricing pages, and sales motions convert measurably better when they speak to a segment instead of everyone.

The discipline: a segment only earns existence if you would treat it differently. Segments without a differentiated action are just charts.

Model 1 — RFM: the classic, adapted for SaaS

RFM comes from retail: score customers on Recency (how recently active), Frequency (how often), and Monetary (how much they pay). Score each dimension 1–5 by quintile, and the combinations describe lifecycle states:

Segment	Profile (R/F/M)	Typical action
Champions	5/5/4-5	Referral asks, case studies, beta access
Loyal but modest	4-5/4-5/1-2	Upsell path, usage-based upgrade nudges
At-risk big accounts	1-2/3-5/4-5	Human outreach this week
Slipping regulars	2-3/3-4/any	Re-engagement, "what changed?" survey
New and promising	5/1-2/any	Onboarding acceleration
Hibernating	1/1-2/1-2	Low-cost win-back, then let go

For SaaS, adapt Monetary to plan value or expansion potential, and Frequency to core-action count rather than logins. The magic of RFM is that it's computable from data you already have, in an afternoon, and immediately produces a worklist — the "at-risk big accounts" cell alone usually pays for the exercise.

Model 2 — Behavioral: segment by what users do

Behavioral segmentation groups users by how they use the product: which features, how deeply, in what patterns. Useful cuts for SaaS:

Activation state: never activated / activated / reached habit (e.g. active 3+ weeks of the last 4).
Feature persona: which core workflow dominates their usage — e.g., in an analytics tool, "dashboard checkers" vs. "report builders" vs. "integrators" (illustrative). Different personas churn for different reasons and want different roadmap items.
Depth: breadth of features touched. Users engaging 3+ core features often retain dramatically better than single-feature users — a pattern worth verifying in your own data, then engineering onboarding toward.
Trajectory: usage rising, stable, or declining over 4 weeks. Declining-usage accounts are your churn early-warning system, weeks before the cancellation.

Behavioral segments drive product decisions the way RFM drives lifecycle marketing. The trap: over-clustering. Fancy clustering algorithms produce statistically real but operationally meaningless groups. Start with hand-defined segments tied to your product's structure; you'll cover most of the value.

Model 3 — Firmographic: segment by who they are

For B2B, the account's shape often predicts behavior before any usage exists: company size, industry, role of the buyer, tech stack, geography. Uses:

ICP validation. Compare NRR and activation across firmographic cells. If 10–50 person agencies retain at 95% NRR while enterprises churn after pilots (illustrative), your ICP has spoken — regardless of what the pitch deck says.
Routing. Firmographics are available at signup, so they can route users instantly: self-serve flow for small teams, sales-assist for 100+ seats.
Pricing design. Persistent willingness-to-pay differences across firmographic segments are the raw material of good packaging.

Sources: signup-form fields (keep them minimal), enrichment providers, and your own inference from email domains. Accept imperfection; 80% coverage is plenty for directional decisions.

What each model feeds

To make the division of labor concrete: RFM segments plug directly into lifecycle messaging — your email tool's audiences should mirror the RFM cells, so "slipping regulars" automatically receive the re-engagement sequence and champions the referral ask. Behavioral personas feed the roadmap and onboarding: each persona gets a tailored first-run path toward its core workflow rather than a generic tour. Firmographic cells feed acquisition targeting and packaging: the segments with the best NRR define where the next ad dollar and the next pricing tier aim. If a segmentation output doesn't plug into one of those three machines — messaging, product, go-to-market — it's analysis for its own sake.

Combining the three

The models answer different questions — who they are (firmographic), what they do (behavioral), and where they stand in the lifecycle (RFM). The practical combinations:

Firmographic × activation state → which ICP segments your onboarding fails.
Behavioral persona × RFM → which usage patterns produce champions vs. hibernators.
Firmographic × NRR → where to point sales and pricing next quarter.

Resist the full three-way cube: with early-stage volumes, most cells will hold six users and pure noise.

A two-week rollout plan

Segmentation projects die from over-scoping. Here's a contained version:

Days 1–2: pick one decision that segmentation would improve right now — usually "which accounts do we contact about churn risk?" or "which signups get the high-touch onboarding?" Scope everything to that decision.

Days 3–5: build the minimal RFM cut in a spreadsheet or your analytics tool. Recency = days since last core action; Frequency = core actions in the last 30 days; Monetary = MRR. Quintile scores, one pivot table.

Days 6–8: eyeball the extreme cells. Read ten actual accounts from "at-risk big accounts" and ten from "champions." Do the segments match reality? If a "champion" turns out to be a bot and an "at-risk" account is just on vacation, refine the definitions now, cheaply.

Days 9–12: run the action. Outreach to the at-risk list, referral asks to champions. Instrument the outcome: replies, saves, referrals generated.

Days 13–14: review and decide whether the segment earns permanence. Illustratively, if outreach to fifteen at-risk accounts saved two ($400 MRR retained for an afternoon of work), the segment has proven its keep — schedule the monthly refresh and only then consider the next model.

One decision, one model, one action, one measurement. Expand only after the first loop pays.

Operational guardrails

Minimum viable segment: ~50–100 users (or ~20 accounts for revenue cuts). Below that, differences are anecdotes.
Every segment gets an owner and an action. Champions → founder sends referral ask. At-risk → success outreach. No action, no segment.
Recompute on a schedule. Segments are states, not identities — users flow between them. Monthly refresh is right for most.
Measure segment transitions, not just sizes. The win isn't "we have 200 champions"; it's "12% of 'new and promising' became champions this quarter, up from 8%."
Write definitions down. "Active" and "big account" drift across tools and teammates; a one-page glossary prevents three versions of the truth.

The bottom line

Segmentation converts averages into decisions: RFM for lifecycle actions, behavioral for product strategy, firmographic for ICP and go-to-market. Keep segments few, sized, owned, and refreshed — and let averages retire.

Growth Pilot's cohort and funnel views let you slice activation, retention, and revenue by the segments you define, so "who is this actually working for?" becomes a filter, not a research project.